![Page 1: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/1.jpg)
Explainable Machine Translation
Aarne RantaDepartment of Computer Science and Engineering
Chalmers University of Technology and University of Gothenburg
and
Digital Grammars AB
Logic and Machine Learning Gothenburg 12-13 June 2017
Explainable Machine Translation with Interlingual Trees as Certificates
Aarne RantaDepartment of Computer Science and Engineering
Chalmers University of Technology and University of Gothenburg
and
Digital Grammars AB
Logic and Machine Learning Gothenburg 12-13 June 2017
The Next Big Disruptive Trend in Business Explainable AI
httpsdisruptionhubcomnext-big-disruptive-trend-business-explainable-ai
18 January 2017
The Next Big Disruptive Trend in Business Explainable AI
httpsdisruptionhubcomnext-big-disruptive-trend-business-explainable-ai
18 January 2017
XAI
input output +explanation
XAI
input output +explanation
httpwwwdarpamilprogramexplainable-artificial-intelligence
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Levels of evidence
formal
total proof of program
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 2: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/2.jpg)
Explainable Machine Translation with Interlingual Trees as Certificates
Aarne RantaDepartment of Computer Science and Engineering
Chalmers University of Technology and University of Gothenburg
and
Digital Grammars AB
Logic and Machine Learning Gothenburg 12-13 June 2017
The Next Big Disruptive Trend in Business Explainable AI
httpsdisruptionhubcomnext-big-disruptive-trend-business-explainable-ai
18 January 2017
The Next Big Disruptive Trend in Business Explainable AI
httpsdisruptionhubcomnext-big-disruptive-trend-business-explainable-ai
18 January 2017
XAI
input output +explanation
XAI
input output +explanation
httpwwwdarpamilprogramexplainable-artificial-intelligence
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Levels of evidence
formal
total proof of program
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 3: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/3.jpg)
The Next Big Disruptive Trend in Business Explainable AI
httpsdisruptionhubcomnext-big-disruptive-trend-business-explainable-ai
18 January 2017
The Next Big Disruptive Trend in Business Explainable AI
httpsdisruptionhubcomnext-big-disruptive-trend-business-explainable-ai
18 January 2017
XAI
input output +explanation
XAI
input output +explanation
httpwwwdarpamilprogramexplainable-artificial-intelligence
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Levels of evidence
formal
total proof of program
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 4: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/4.jpg)
The Next Big Disruptive Trend in Business Explainable AI
httpsdisruptionhubcomnext-big-disruptive-trend-business-explainable-ai
18 January 2017
XAI
input output +explanation
XAI
input output +explanation
httpwwwdarpamilprogramexplainable-artificial-intelligence
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Levels of evidence
formal
total proof of program
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 5: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/5.jpg)
XAI
input output +explanation
XAI
input output +explanation
httpwwwdarpamilprogramexplainable-artificial-intelligence
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Levels of evidence
formal
total proof of program
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 6: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/6.jpg)
XAI
input output +explanation
httpwwwdarpamilprogramexplainable-artificial-intelligence
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Levels of evidence
formal
total proof of program
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 7: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/7.jpg)
httpwwwdarpamilprogramexplainable-artificial-intelligence
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Levels of evidence
formal
total proof of program
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 8: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/8.jpg)
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Levels of evidence
formal
total proof of program
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 9: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/9.jpg)
Kurt Mehlhorn Certifying Algorithmshttpspeoplempi-infmpgde~mehlhornftpSAPJuly2014pdf
Levels of evidence
formal
total proof of program
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 10: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/10.jpg)
Levels of evidence
formal
total proof of program
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 11: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/11.jpg)
Levels of evidence
formal
total proof of program
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 12: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/12.jpg)
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 13: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/13.jpg)
Levels of evidence
formal informal
total proof of program correctness by construction
individual certificate of instance explanation of instance
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 14: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/14.jpg)
Evidence for SMT
SMT = Statistical Machine Translation
- glue together segments from aligned texts
Informal evidence phrase alignments
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 15: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/15.jpg)
httpstranslategooglecom 13 mars 2017
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 16: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/16.jpg)
This big car is yellow Denna stora bil aumlr gul
This house is clean Detta hus aumlr rent
This big house is yellow Denna stora hus aumlr gul
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 17: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/17.jpg)
Evidence for NMT
NMT = Neural Machine Translation
- end-to-end string conversion via a neural network
Individual explanations word vector ldquointerlinguardquo
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 18: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/18.jpg)
httpsgab41lab41orgcan-word-vectors-help-predict-whether-your-chinese-tweet-gets-censored-711e7682d12f
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 19: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/19.jpg)
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 20: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/20.jpg)
httpstranslategooglecom 5 May 2017
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 21: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/21.jpg)
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 22: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/22.jpg)
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 23: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/23.jpg)
httpstranslategooglecom 5 May 2017
Possible evidence translation to some language you know
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 24: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/24.jpg)
From SMT to NMTBLEU (max 10)
Fluency (max 60)
SMT NMT
037 041
SMT NMT human
387 444 482
Wu amp al Bridging the Gap Between Machine and Human Translation 2016
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 25: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/25.jpg)
From SMT to NMT
+ improved average scores+ increased fluency
- harder to predict- harder to explain
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 26: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/26.jpg)
Nguyen amp al Deep Neural Networks are Easily Fooled CVPRrsquo15 2015
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 27: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/27.jpg)
httpstranslategooglecom 5 maj 2017
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 28: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/28.jpg)
XMT our proposal
- Explainable Machine Translation
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 29: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/29.jpg)
What to verify in translation
1 The output is a valid expression of the target language
2 The output has the same meaning as the input
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 30: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/30.jpg)
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 31: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/31.jpg)
GF = Grammatical FrameworkLF = Logical Framework = framework for defining logics
GF = LF + linearization = framework for defining grammars
interlingua = abstract syntax = type theoretical logic
abstract syntax
linearizationEnglish
Finnish
AR Grammatical Framework CSLI 2011
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 32: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/32.jpg)
The Vauquois triangle answer
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 33: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/33.jpg)
Parsing in GFReverse of linearization
abstract syntax
parsing
linearizationEnglish
Finnish
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 34: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/34.jpg)
K Angelov Incremental Parsing with Parallel Multiple Context-Free Grammars EACL 2009
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 35: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/35.jpg)
The Vauquois triangle answer variant 1
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
provable by reversibility of linearization
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 36: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/36.jpg)
Problems with variant 1Ambiguity
Linearization is many-to-1
rarr Parsing is 1-to-many
Incompleteness
The grammar doesnrsquot cover all input
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 37: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/37.jpg)
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 38: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/38.jpg)
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 39: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/39.jpg)
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 40: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/40.jpg)
The Vauquois triangle answer variant 2
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 41: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/41.jpg)
XMT
source target ++explanation
morphology
semantic interlingua
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 42: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/42.jpg)
XMT
source target ++explanation(semantic interlingua)
morpho- logy
lineari-zation
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 43: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/43.jpg)
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 44: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/44.jpg)
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 45: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/45.jpg)
httpwordnetwebprincetonedu
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 46: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/46.jpg)
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
UD parsing
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 47: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/47.jpg)
UD = Universal DependenciesDependency tree labelled arcs between words
Universal same labels in different languages
Parsing machine-learned from treebanks
treebanks training
English Finnish hellip
parsers
UD trees
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 48: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/48.jpg)
Languages in UD and GFBasque
Belarusian BuryatCoptic CroatianCzech Galician
HungarianIndonesian IrishKazakh Korean
Kurmanji LithuanianNorthSami
OldChurchSlavonicPortuguese Sanskrit
Slovak TamilUkranian
UpperSorbian UyghurVietnamese
Arabic Bulgarian
Catalan ChineseDanish Dutch English
Estonian Finnish French German GothicGreek(AncientModern)
Hebrew Hindi Italian Japanese
Latin Latvian MalteseNorwegian(bokmaringlnynorsk)
Persian Polish Romanian Russian
Slovenian Spanish SwedishThai Turkish Urdu
AfrikaansAmharicIcelandicMongolianNepaliPunjabiSindhiSwahili
httpwwwgrammaticalframeworkorglibdocstatushtmlhttpuniversaldependenciesorg
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 49: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/49.jpg)
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 50: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/50.jpg)
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
the black cat sees us today
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 51: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/51.jpg)
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
Kolachina amp Ranta From Abstract Syntax to Universal Dependencies LiLT 2016
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 52: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/52.jpg)
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 53: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/53.jpg)
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 54: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/54.jpg)
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
det
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 55: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/55.jpg)
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
advmoddet
amod
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 56: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/56.jpg)
abstract syntax
PredVP NP -gt VP -gt Cl
ComplV2 V2 -gt NP -gt VP
AdvVP VP -gt Adv -gt VP
DetCN Det -gt CN -gt NP
ModCN AP -gt CN -gt CN
UseN N -gt CN
UsePron Pron -gt NP
PositA A -gt AP
dependency configuration
nsubj head
head dobj
head advmod
det head
amod head
head
head
head
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 57: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/57.jpg)
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 58: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/58.jpg)
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 59: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/59.jpg)
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 60: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/60.jpg)
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 61: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/61.jpg)
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 62: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/62.jpg)
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 63: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/63.jpg)
nsubj
dobj
advmoddet
amod
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 64: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/64.jpg)
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 65: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/65.jpg)
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 66: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/66.jpg)
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 67: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/67.jpg)
nsubj
dobj
advmoddet
amod
the_Det
black_A cat_N
see_V2
we_Pron
today_Adv
abstract syntax category configuration
Det DETA ADJN NOUNV2 VERBPron PRONAdv ADV
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 68: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/68.jpg)
nsubj
dobj
advmoddet
amod
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 69: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/69.jpg)
Example pipeline 1
English UD tree GF tree FrenchUD parser ud2gf GF linearizer
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 70: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/70.jpg)
if a man owns a donkey it beats he
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 71: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/71.jpg)
PARSER OUTPUT IN CONLL FORMAT
1 if if SCONJ SCONJ _ 4 mark _ _2 a a DET DET Definite=Ind|PronType=Art 3 det _ _3 man man NOUN NOUN Number=Sing 4 nsubj _ _4 owns own VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl _ _5 a a DET DET Definite=Ind|PronType=Art 6 det _ _6 donkey donkey NOUN NOUN Number=Sing 4 dobj _ _7 it it PRON PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 8 nsubj _ _8 beats beat VERB VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 0 root _ _9 he he PRON PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 8 dobj _ _
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 72: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/72.jpg)
STRUCTURED TREE
root beat VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 8 advcl own VERB Mood=Ind|Number=Sing|Person=3|Tense=Pres|VerbForm=Fin 4 mark if SCONJ _ 1 nsubj man NOUN Number=Sing 3 det a DET Definite=Ind|PronType=Art 2 dobj donkey NOUN Number=Sing 6 det a DET Definite=Ind|PronType=Art 5 nsubj it PRON Case=Nom|Gender=Neut|Number=Sing|Person=3|PronType=Prs 7 dobj he PRON Case=Nom|Gender=Masc|Number=Sing|Person=3|PronType=Prs 9
Ranta amp Kolachina From Universal Dependencies to Abstract Syntax UD Workshop 2017
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 73: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/73.jpg)
LEXICALLY ANNOTATED TREE
root VERB beat_V2 V2 [beat_V V] (8) 8 advcl VERB own_V2 V2 [] (4) 4 mark SCONJ if_Subj Subj [] (1) 1 nsubj NOUN man_N N [] (3) 3 det DET IndefArt Quant [] (2) 2 dobj NOUN donkey_N N [] (6) 6 det DET IndefArt Quant [] (5) 5 nsubj PRON it Cleft_ [it_Pron Pron] (7) 7 dobj PRON he_Pron Pron [] (9) 9
GF lexicon
fun beat_V2 V2lin beat_V2 = mkV2 IrregEngbeat_V
fun own_V2 V2lin own_V2 = mkV2 ldquoownrdquo
fun man_N Nlin man_N = mkN ldquomanrdquo ldquomenrdquo
fun donkey_N Nlin donkey_N = mkN ldquodonkeyrdquo
fun he_Pron Pronfun it_Pron Pronfun Cleft_ NP -gt RS -gt Clfun IndefArt Quantfun if_Subj Subj
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 74: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/74.jpg)
A part of GF Resource Grammar Abstract Syntax
fun PredVP NP -gt VP -gt Cl ComplV2 V2 -gt NP -gt VP DetCN Det -gt CN -gt NP DetQuant Quant -gt Num -gt Det AdvS Adv -gt S -gt S SubjS Subj -gt S -gt Adv UseCl Temp -gt Pol -gt Cl -gt S UsePron Pron -gt NP UseN N -gt CN
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 75: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/75.jpg)
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 76: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/76.jpg)
Dependency configurations for abstract syntax
fun PredVP NP -gt VP -gt Cl -- nsubj head ComplV2 V2 -gt NP -gt VP -- head dobj DetCN Det -gt CN -gt NP -- det head DetQuant Quant -gt Num -gt Det -- head [nummod] AdvS Adv -gt S -gt S -- advcl head SubjS Subj -gt S -gt Adv -- mark head UseCl Temp -gt Pol -gt Cl -gt S -- [aux] [neg] head UsePron Pron -gt NP -- head UseN N -gt CN -- head
Helper functions
DetQuantSg_ Quant -gt Det = q -gt DetQuant q NumSg UseClPresPos_ Cl -gt S = cl -gt UseCl Pres Pos cl
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 77: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/77.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 78: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/78.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 79: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/79.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 80: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/80.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 det DET IndefArt Quant 5 nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 81: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/81.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 82: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/82.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 83: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/83.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 84: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/84.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 85: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/85.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 86: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/86.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 87: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/87.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 88: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/88.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 89: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/89.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 90: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/90.jpg)
PredVP NP -gt VP -gt Cl -- nsubj headComplV2 V2 -gt NP -gt VP -- head dobjDetCN Det -gt CN -gt NP -- det headAdvS Adv -gt S -gt S -- advcl headSubjS Subj -gt S -gt Adv -- mark headUsePron Pron -gt NPUseN N -gt CNDetQuantSg_ Quant -gt DetUseClPrPos_ Cl -gt S
TRAVERSING THE TREE (AdvS 4 8)root VERB beat_V2 V2 [beat_V V] 8 (ComplV2 8 9) (PredVP 7 8) (UseClPrPos_ 8) advcl VERB own_V2 V2 4 (ComplV2 4 6) (PredVP 3 4) (UseClPrPos_ 4) (AdvS 1 4) mark SCONJ if_Subj Subj 1 nsubj NOUN man_N N 3 (UseN 3) (DetCN 2 3) det DET IndefArt Quant 2 (DetQuantSg_ 2) dobj NOUN donkey_N N 6 (UseN 6) (DetCN 5 6) det DET IndefArt Quant 5 (DetQuantSg_ 5) nsubj PRON it Cleft_ [it_Pron Pron] 7 (UsePron 7) dobj PRON he_Pron Pron 9 (UsePron 9)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 91: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/91.jpg)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseNdonkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 92: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/92.jpg)
si un homme possegravede un acircne il le bat
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 93: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/93.jpg)
si un homme possegravede un acircne il le bat
if a man owns a donkey it beats him if a man owns a donkey he beats it
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 94: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/94.jpg)
Example pipeline 2
English UD tree GF tree TT formula
UD parser ud2gf gf2tt
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 95: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/95.jpg)
echo ldquoif a man owns a donkey it beat herdquo | udjpipescriptspipelinesh -l en ud2gf -lEng -t10000 -k3000 -a1 -g1 -Dt -CUDTranslatelabelsUDTranslateEnglabels runghc TTGhs
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 96: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/96.jpg)
iS GS -gt Prop
iS s = case s of
GUseCl _ pol cl -gt iPol pol (iCl cl)
GAdvS (GSubjS Gif_Subj a) b -gt Pi (iS a) (x -gt iS b) --- non-compositional
hellipiCl GCl -gt Prop
iCl s = case s of
GPredVP np vp -gt iNP np (iVP vp)
hellipiVP GVP -gt Ind -gt Prop
iVP vp x = case vp of
GComplV2 v np -gt iNP np (y -gt iV2 v x y)
hellipiNP GNP -gt (Ind -gt Prop) -gt Prop
iNP np p = case np of
GDetCN (GDetQuant GDefArt _) cn -gt p (Def (iCN cn) [])
GDetCN det cn -gt iDet det (iCN cn) p
hellipiDet GDet -gt Prop -gt (Ind -gt Prop) -gt Prop
iDet det t p = case det of
GsomeSg_Det -gt Sigma t p
every_Det -gt Pi t p
GDetQuant GIndefArt _ -gt Sigma t p --- non-compositional
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 97: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/97.jpg)
(Πz (Σx man_N)(Σy donkey_N)own_V2(xy)) beat_V2(p(q(z))p(z))
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 98: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/98.jpg)
Example pipeline 3
English UD tree GF tree French
TT formula
GF treeUD parser ud2gf
gf2tt
(gf+tt)2gf
GF linearizer
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 99: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/99.jpg)
AR Type Theoretical Grammar OUP 1994
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 100: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/100.jpg)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (UsePron it_Pron) (ComplV2 beat_V2 (UsePron he_Pron)))))
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 101: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/101.jpg)
GF ABSTRACT SYNTAX TREE
(AdvS (SubjS if_Subj (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant IndefArt NumSg) (UseN man_N)) (ComplV2 own_V2 (DetCN (DetQuant IndefArt NumSg) (UseN donkey_N)))))) (UseCl (TTAnt TPres ASimul) PPos (PredVP (DetCN (DetQuant DefArt NumSg) (UseN donkey_N)) (ComplV2 beat_V2 (UsePron he_Pron)))))
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 102: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/102.jpg)
si un homme possegravede un acircne lrsquoacircne le bat ldquoif a man owns a donkey the donkey beats himrdquo
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 103: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/103.jpg)
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 104: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/104.jpg)
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 105: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/105.jpg)
Research goals
Linguistics- What are the structures of languageStrong AI- Can machines learn to perform like humansEngineering- How best to build systems that work
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 106: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/106.jpg)
Donrsquot guess if you know
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 107: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/107.jpg)
Kyunghyun Cho Keynote on NMT NODALIDA 2017
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 108: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/108.jpg)
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 109: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/109.jpg)
ResourceDemo num (pot3plus (pot1as2 (pot0as1 (pot0 n7))) (pot2plus (pot0 n2) (pot1plus n5 (pot0 n4))))ResourceDemoAfr sevenduisend tweehonderdvierenvyftigResourceDemoBul седем хиляди двеста петдесет и четиримаResourceDemoCat set mil dos -cents cinquanta- quatreResourceDemoChi 七 千 两 百 五 十 四ResourceDemoDan syv tusind og to hundrede og fire og halvtredsResourceDemoDut zevenduizend tweehonderdvierenvijftigResourceDemoEng seven thousand two hundred and fifty-fourResourceDemoEst seitse tuhat kakssada viiskuumlmmend neliResourceDemoFin seitsemaumlntuhatta kaksisataaviisikymmentaumlneljaumlResourceDemoFre sept mille deux cent cinquante-quatreResourceDemoGer siebentausend zweihundertvierundfuumlnfzigResourceDemoGre εφτά χιλιάδες διακόσιοι πενήντα τέσσερειςResourceDemoHin सात हज़ार दो सौ चwwनResourceDemoIce sjouml thornuacutesund tvouml hundraeth fimmtugasti og fjoacuterethiResourceDemoIta settemila e duecentocinquantaquattroResourceDemoJpn 七 千 二 百 五 十 四ResourceDemoLav septiņi tūkstoši divi simti piecdesmit četriResourceDemoMlt sebat elef u mitejn u erbgħa u ħamsinResourceDemoMon долоон мянга хоёр зуун тавин дөрөвResourceDemoNep सात हजार दई सय चवननResourceDemoNno sju tusen og to hundre og femti fireResourceDemoNor sju tusen og to hundre og femti fireResourceDemoPes ھفت ھزار و دویست و پنجاه و چھارResourceDemoPnb ست ہزار دو سو چوتنجاResourceDemoPol siedem tysięcy dwieście pięćdziesiąt czteryResourceDemoRon şapte mii două sute cincizeci şi patruResourceDemoRus семь тысяч двести пятьдесят четыреResourceDemoSnd ست ھزار ٻ سو چوونجاھResourceDemoSpa siete mil doscientos cincuenta y cuatroResourceDemoSwe sjutusen tvaringhundra femtiofyraResourceDemoTha เจด พน สอง รอย หา สบ สResourceDemoUrd سات ہزار دو سو چوون
Hammarstroumlm amp Ranta Cardinal Numerals Revisited in GF 2004
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 110: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/110.jpg)
Descartes letter to Mersenne 1629
In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 111: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/111.jpg)
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 112: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/112.jpg)
Descartes letter to Mersenne 1629In a single day one can learn to name every one of the infinite series of numbers and thus to write infinitely many different words in an unknown language The same could be done for all the other words necessary to express all the other things which fall within the purview of the human mind
the discovery of such a language depends upon the true philosophy For without that philosophy it is impossible to number and order all the thoughts of men or even to separate them out into clear and simple thoughts
But do not hope ever to see such a language in use For that the order of nature would have to change so that the world turned into a terrestrial paradise and that is too much to suggest outside of fairyland
httpwwwautodidactprojectorgotherdescartes-lg1html
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 113: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/113.jpg)
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by constructionwell
understood
uncertain
highly uncertain
incomplete
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 114: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/114.jpg)
semantic interlingua
lexical analysis
syntactic parsing
semantic interpretation
syntactic form generation
lexical surface generation
source target
correct by construction
correct by construction
well understood
uncertain
highly uncertain
incomplete
highly uncertain
uncertain
lexical transfer
syntactic transfer
character transfer
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 115: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/115.jpg)
semantic interlingua
source target
complete
incomplete
highly incomplete
chunk interlingua
syntactic interlingua
correct by construction
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 116: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/116.jpg)
precision 100(BLEU)
20
100 1000 1000000 concepts coverage
producer tasks
consumer tasks
Producer vs consumer task
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 117: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/117.jpg)
TitleParagraph DefinitionTitleDefPredParagraph type_Sort A_Var contractible_Pred (ExistCalledProp a_Var (ExpSort (VarExp A_Var)) (FunInd centre_of_contraction_Fun) (ForAllProp (BaseVar x_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (VarExp a_Var) (VarExp x_Var)))))FormatParagraph EmptyLineFormatTitleParagraph DefinitionTitleDefPredParagraph (mapSort (mapExp (VarExp A_Var) (VarExp B_Var))) f_Var equivalence_Pred (ForAllProp (BaseVar y_Var) (ExpSort (VarExp B_Var)) (PredProp contractible_Pred (AliasInd (AppFunItInd fiber_Fun) (FunInd (ExpFun (ComprehensionExp x_Var (VarExp A_Var) (equalExp (AppExp f_Var (VarExp x_Var)) (VarExp y_Var))))))))DefPropParagraph (ExpProp (equivalenceExp (VarExp A_Var) (VarExp B_Var))) (ExistSortProp (equivalenceSort (mapExp (VarExp A_Var) (VarExp B_Var))))FormatParagraph EmptyLineFormatTitleParagraph LemmaTitleTheoremParagraph (ForAllProp (BaseVar A_Var) type_Sort (PredProp equivalence_Pred (AliasInd (FunInd identity_map_Fun) (FunInd (ExpFun (DefExp (identityMapExp (VarExp A_Var)) (TypedExp (BaseExp (lambdaExp x_Var (VarExp A_Var) (VarExp x_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))))FormatParagraph EmptyLineFormatTitleParagraph ProofTitleAssumptionParagraph (ConsAssumption (ForAssumption y_Var (ExpSort (VarExp A_Var)) (LetAssumption (FunInd (ExpFun (DefExp (fiberExp (VarExp y_Var) (VarExp A_Var)) (ComprehensionExp x_Var (VarExp A_Var) (equalExp (VarExp x_Var) (VarExp y_Var)))))) (AppFunItInd (fiberWrt_Fun (FunInd (ExpFun (identityMapExp (VarExp A_Var)))))))) (BaseAssumption (LetExpAssumption (barExp (VarExp y_Var)) (TypedExp (BaseExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var)))) (fiberExp (VarExp y_Var) (VarExp A_Var))))))ConclusionParagraph (AsConclusion (ForAllProp (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ExpProp (equalExp (pairExp (VarExp y_Var) (reflexivityExp (VarExp A_Var) (VarExp y_Var))) (VarExp y_Var)))) (ApplyLabelConclusion id_induction_Label (ConsInd (FunInd (ExpFun (VarExp y_Var))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp x_Var)) (VarExp A_Var)))) (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp z_Var)) (idPropExp (VarExp x_Var) (VarExp y_Var))))) BaseInd))) (DisplayExpProp (equalExp (pairExp (VarExp x_Var) (VarExp z_Var)) (VarExp y_Var)))))ConclusionSoThatParagraph (ForConclusion (BaseVar y_Var) (ExpSort (VarExp A_Var)) (ApplyLabelConclusion sigma_elimination_Label (ConsInd (FunInd (ExpFun (TypedExp (BaseExp (VarExp u_Var)) (fiberExp (VarExp y_Var) (VarExp A_Var))))) BaseInd) (ExpProp (equalExp (VarExp u_Var) (VarExp y_Var))))) (PredProp contractible_Pred (FunInd (ExpFun (fiberExp (VarExp y_Var) (VarExp A_Var)))))ConclusionParagraph (PropConclusion (PredProp equivalence_Pred (FunInd (ExpFun (TypedExp (BaseExp (identityMapExp (VarExp A_Var))) (mapExp (VarExp A_Var) (VarExp A_Var)))))))QEDParagraph
httpsgithubcomGrammaticalFrameworkgf-contribtreemasterhomotopy-typetheory
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 118: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/118.jpg)
chunk interlingua
syntactic interlingua
semantic interlingua
Cf Vauquois (1968)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 119: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/119.jpg)
GF Offline Translatorhttpsplaygooglecomstoreappsdetailsid=orggrammaticalframeworkuiandroid
httpsitunesapplecomusappgf-offline-translatorid1023328422mt=8
K Angelov B Bringert amp A Ranta Speech-enabled hybrid multilingual translation for mobile devices EACL 2014
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 120: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/120.jpg)
Quality degradation precision 95 75 50 35 25
H-BLEU 100 1000 1000000 concepts coverage
GF semantic
NMT SMT
GF syntactic GF chunk
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 121: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/121.jpg)
STRING Fast and friendly service they know my order when I walk in the door
root NOUN service_N N [] (4) 4 amod ADJ fast_A A [] (1) 1 cc CONJ and Conjand_ [and_Conj Conj] (2) 2 conj ADJ friendly_A A [] (3) 3 punct PUNCT Comma_ [] (5) 5 parataxis VERB know_VQ VQ [know_VS VS know_V2 V2 know_V V] (7) 7 nsubj PRON they_Pron Pron [theyFem_Pron Pron] (6) 6 dobj NOUN order_N N [] (9) 9 nmodposs PRON i_Pron Pron [] (8) 8 advcl VERB walk_V2 V2 [walk_V V] (12) 12 mark ADV when_Subj Subj [when_IAdv IAdv] (10) 10 nsubj PRON i_Pron Pron [iFem_Pron Pron] (11) 11 nmod NOUN door_N N [] (15) 15 case ADP in_Prep Prep [] (13) 13 det DET DefArt Quant [] (14) 14 punct PUNCT StringPN PN [StringPunct Punct] (16) 16
Eng fast and friendly service [ they know my order when I walk in the door ]Fin nopea ja ystaumlvaumlllinen palvelu [ he tuntevat minun jaumlrjestykseni kun minauml kaumlvelen ovessa ]Swe snabb och vaumlnlig tjaumlnst [ de kaumlnner min ordning naumlr jag garingr i doumlrren ]
PARSED 1616 WITHOUT BACKUP 616
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home
![Page 122: Chalmers University of Technology and University of ...old-darcs.grammaticalframework.org/~aarne/xmt-2017.pdf · Sindhi Swahili ... if a man owns a donkey it beats he. PARSER OUTPUT](https://reader030.vdocuments.site/reader030/viewer/2022020315/5ac133de7f8b9a357e8c6029/html5/thumbnails/122.jpg)
UUUUU
XMT
source target ++explanation(semantic interlingua)
morpho- logy
Checker C = back-linearization
lineari-zation
parsing and
semantic interpretation
Take home