lefrancoisgandon@msw11: ulis
DESCRIPTION
Slides of the second paper on the ULiS project, availiable at http://maxime-lefrancois.info/Publications We are interested in bridging the world of natural language and the world of the semantic web in particular to support multilingual access to the web of data. In this paper we introduce the ULiS project, that aims at designing a pivot-based NLP technique called Universal Linguistic System, 100% using the semantic web formalisms, and being compliant with the Meaning-Text theory. Through the ULiS, a user could interact with an interlingual knowledge base (IKB) in controlled natural language. Linguistic resources themselves are part of a specific IKB: The Universal Lexical Knowledge base (ULK), so that actors may enhance their controlled natural language, through requests in controlled natural language. We describe a basic interaction scenario at the system level, and provide an overview of the architecture of ULiS. We then introduce the core of the ULiS: the interlingual lexical ontology (ILexicOn), in which each interlingual lexical unit class (ILUc) supports the projection of its semantic decomposition on itself. We validate our model with a standalone ILexicOn, and introduce and explain a concise human-readable notation for it.TRANSCRIPT
Maxime Lefrançois & Fabien GandonEdelweiss – INRIA Sophia-Antipolis – France{Maxime.Lefrancois|Fabien.Gandon}@inria.fr
MSW 2011
ULiS: An Expert System on Linguisticsto Support Multilingual Management
of Interlingual Semantic Web Knowledge bases
MSW 2011Lefrançois & Gandon, ULiS. - 2
multilingual access to the web of dataThe ULiS project
?
>31 billion RDF triples295 data setsInterlinked by >503 million RDF links
3000 < natural languages < 7000
Lefrançois & Gandon, ULiS.
OUTLINE
The Meaning-Text Theory
The ULiS project: multilingual access to web of data
The ILexicOn : defining interlingual lexical units
MSW 2011 - 3
Proposal 2: A formal way to represent lexicographic definitions
Proposal 1: An architecture that unlocks motivating scenarios
1. The Meaning-Text Theory
MSW 2011Lefrançois & Gandon, ULiS.
The Meaning-Text Theory
MSW 2011 - 5
A theoretical linguistic framework to construct models of natural language
Mel’čuk et. al., late 70’s > today
Lefrançois & Gandon, ULiS.
MSW 2011Lefrançois & Gandon, ULiS. - 6
The Meaning-Text Theory
SSynRs
DMorphRs
SMorphRs
DPhonRs
SPhonRs
DSynRs
SemRs
Systems of explicit rulesthat express correspondences
between meaning and text
7 levels of linguistic representationFor a set of synonymous utterances
MSW 2011Lefrançois & Gandon, ULiS. - 7
The Meaning-Text Theory
Description of theExplanatory
CombinatorialDictionary
SSynRs
DMorphRs
SMorphRs
DPhonRs
SPhonRs
DSynRs
SemRs
Explanatory Combinatorial
Dictionary
Explantory: Formal semantic explanations
Combinatorial:Lexical combinations
The masterpiece: The Explanatory Combinatorial Dictionary
Meta- Ontology Data
MSW 2011Lefrançois & Gandon, ULiS. - 8
Structure of the ECD entry
Semantic zone: (formal?) (text or SemR?) lexical definition of lexical unitsSemantic actants: arguments of the semantic predicatedenotational vs. connotational semantic elements (ex: EN hot air, FR wind, RU water).
Phonological/Graphematic zone: orthograph + prosody
The cooccurrence zone
MSW 2011Lefrançois & Gandon, ULiS. - 9
Structure of the ECD entry - The cooccurrence zone
Morphological sub-zone: inflections, derivations.
Pragmatic sub-zonereal-life situations where expressions are appropriate.
Stylistic sub-zoneUsage labels
Lexical sub-zonesemantic derivations and collocations
Syntactic sub-zone: Government pattern
Correspondance semantic – syntax (based on Dependency grammars)ways to express a semantic predicate in the language
How the lexical unit participates in specific constructions as a dependent
MSW 2011Lefrançois & Gandon, ULiS. - 10
Structure of the ECD entry - Examples of lexical functions
lexical functions :
paradigmaticSyn(telephoneV) = phoneV
AntiConv21(preced) = followGener(republic) = stateFigur(rain) = curtain (a curtain of rain)S0 (analyse) = analysisA0(city) = urbanV0(analysis) = analyseAdv0(slow) = slowly S1(teach) = teacher ; S2(teach) = subject ; S3(teach) = pupil;… 19 in total
MSW 2011Lefrançois & Gandon, ULiS. - 11
Structure of the ECD entry - Examples of lexical functions
lexical functions :
syntagmaticMagn(laughV) = one’s head offBon(service) = first-classLocin(height) = at [a height of…]Oper1(supportN) = [to] lend [~ to N]
Oper2(supportN) = [to] receive [~ to N]Func1(blownN) = comes [from ~]… 34 in total
+ non-standard lexical functionsWithNoSugar(coffee) = black
2. The ULiS project:Multilingual access to the web of data
?MSW 2011Lefrançois & Gandon, ULiS.
Universal Networking Language
MSW 2011Lefrançois & Gandon, ULiS. - 13
A pivot-based NLP technique
+MTT?+RDF?
ConvertDeconvert
Interlanguage
MSW 2011Lefrançois & Gandon, ULiS. - 14
ConvertDeconvert
Interlanguage
a Universal Linguistic System
The ULiS project
Lefrançois & Gandon, ULiS.
The ULiS project
MSW 2011 - 15
Meaning-Text Theory
Semantic Web formalisms
Pivot-based NLP technique
ULiS
to redesign
compliant with
100% using
a Universal Linguistic System
2.The ULiS projectScenario
MSW 2011Lefrançois & Gandon, ULiS.
Lefrançois & Gandon, ULiS.
John killed Mary. John a tué Mary.
John01 kill@past Mary01.John01 kill@past Mary01.
John killed Mary
John01 tuer@past Mary01.John01 tuer@past Mary01
John a tué Mary
InputTEXT
IR RDF
Output1TEXT
SREN RDF SRFR
RDF
MSW 2011 - 17
The RDF-World
RDF situational representations
RDF interlingual representations
The ULiS project - scenario- Machine Translation
--
Lefrançois & Gandon, ULiS. MSW 2011 - 18
The RDF-World
RDF situational representations
RDF interlingual representations
John killed Mary. John a tué Mary.
John01 kill@past Mary01.John01 kill@past Mary01.
John killed Mary
John01 tuer@past Mary01.John01 tuer@past Mary01
John a tué Mary
The ULiS project - scenario- Machine Translation
--
Lefrançois & Gandon, ULiS.
IDBpedia
SELECT ?personWHERE { ?person ikb:kill ikb:Mary01. }
ikb:John01
Who killed Mary?
Who kill@past@? Mary01Who kill@past@? Mary01
Who killed Mary ?
John killed Mary.
John01 kill@past Mary01.John01 kill@past Mary01.
John killed MarySREN
RDF
SPARQLRDF + X RDF X RDFIKBRDF
SREN RDF
InputTEXT
IR RDF IR RDF
Output2TEXT
MSW 2011 - 19
The RDF-World
RDFOutput
SPARQL Request
RDF situational representations
RDF interlingual representations
The ULiS project - scenario
- Management of Interlingual Knowledge Bases- Machine Translation
-
Lefrançois & Gandon, ULiS. MSW 2011 - 20
SELECT ?personWHERE { ?person ikb:kill ikb:Mary01. }
ikb:John01IDBpedia
The RDF-World
RDFOutput
SPARQL Request
RDF situational representations
RDF interlingual representations
Who killed Mary?
Who kill@past@? Mary01Who kill@past@? Mary01
Who killed Mary ?
John killed Mary.
John01 kill@past Mary01.John01 kill@past Mary01.
John killed Mary
The ULiS project - scenario
- Management of Interlingual Knowledge Bases- Machine Translation
-
Lefrançois & Gandon, ULiS. MSW 2011 - 21
- Management of Interlingual Knowledge Bases- Machine Translation
- Management of the Universal Linguistic Knowledge base
SPARQLRDF + X RDF X RDF
The RDF-World
RDFOutput
SPARQL Request
RDF situational representations
RDF interlingual representations
InputTEXT
IR RDF
SRFR RDF
Output2TEXT
SRFR RDF
IR RDF
Output1TEXT
SRFR RDF
IKBRDFULKRDF
The ULiS project - scenario
2.The ULiS projectArchitecture
MSW 2011Lefrançois & Gandon, ULiS.
Lefrançois & Gandon, ULiS. MSW 2011 - 23
The ULiS project - architecture
Explanatory Combinatorial
Dictionary
Description of theExplanatory
CombinatorialDictionary
SSynRs
DMorphRs
SMorphRs
DPhonRs
SPhonRs
DSynRs
SemRs
Meta- Ontology Data
Lefrançois & Gandon, ULiS. MSW 2011 - 24
The ULiS project - architecture
Explanatory Combinatorial
Dictionary
Description of theExplanatory
CombinatorialDictionary
SSynRs
DMorphRs
SMorphRs
DPhonRs
SPhonRs
DSynRs
SemRs
X RDFInterlingualKnowledgeBase
Meta- Ontology Data
Lefrançois & Gandon, ULiS.
X RDFInterlingualKnowledgeBase
MSW 2011 - 25
INTERLINGUAL
NOT INTERLINGUAL = SITUATIONAL
The ULiS project - architecture
+RDF!
Lefrançois & Gandon, ULiS.
X RDFInterlingualKnowledgeBase
IRs RDF
MSW 2011 - 26
SRs RDF
SSynRs
DMorphRs
SMorphRs
DPhonRs
SPhonRs
DSynRs
SLexiMOn
« SituationalLexical
Meta-Ontology »
SLexicOn
Other features of the ECD
ILexiMOn
« Interlingual Lexical
Meta-Ontology »
ILexicOn
pure interlingual features of the ECD
The ULiS project - architecture
Lefrançois & Gandon, ULiS.
X RDF
(ex:Maxime,ikb:isFrom,geo:France)
InterlingualKnowledgeBase
(ikb:isFrom,rdfs:range,ikb:Country)(ikb:isFrom,rdfs:domain,ikb:Person)
IRs RDF
ex:John01
MSW 2011 - 27
SRs RDF
SSynRs
DMorphRs
SMorphRs
DPhonRs
SPhonRs
DSynRs
SLexiMOn
sleximon:SLexicalUnit
SLexicOn
enlexicon:Personeslexicon:Personafrlexicon:Personne
ILexiMOn
ileximon:ILexicalUnit
ILexicOn
ilexicon:Person
The ULiS project - architecture
Lefrançois & Gandon, ULiS.
X RDFInterlingualKnowledgeBase
IRs RDF
MSW 2011 - 28
SRs RDF
SSynRs
DMorphRs
SMorphRs
DPhonRs
SPhonRs
DSynRs
SLexiMOn
« SituationalLexical
Meta-Ontology »
SLexicOn
Other features of the ECD
ILexiMOn
« Interlingual Lexical
Meta-Ontology »
ILexicOn
pure interlingual features of the ECD
The ULiS project - architecture
+ Links+ Transformation rules
IKB = KB + anchors + transformation rules
Lefrançois & Gandon, ULiS.
X RDFInterlingualKnowledgeBase
IRs RDF
MSW 2011 - 29
SRs RDF
SSynRs
DMorphRs
SMorphRs
DPhonRs
SPhonRs
DSynRs
SLexiMOn
« SituationalLexical
Meta-Ontology »
SLexicOn
Other features of the ECD
The ULiS project – Done / To do ?
+ Links+ Transformation rules
IKB = KB + anchors + transformation rules
ILexiMOn
« Interlingual Lexical
Meta-Ontology »
ILexicOn
pure interlingual features of the ECD
Formal lexicographic definitions of interlingual lexical units…
in RDF!
(I)RDF
(I)SPIN
(I)RDF-S
(I)OWL
(I)ILexiMOn (I)SKOS, (I)FOAF, …
(I)SLexiMOn
(I)SLexicOn
(I)ISlexicOn
The ULiS project – Bricks
MSW 2011 - 30Lefrançois & Gandon, ULiS.
3. The ILexicOn:defining interlingual lexical units
MSW 2011Lefrançois & Gandon, ULiS.
With the Semantic Web formalisms, We designed a simple ILexiMOn…
Interlingual Lexical Units Classes may be formally defined in the ILexicOn…
…by supporting the projection of their semantic decomposition
on themselves
Lefrançois & Gandon, ILexicOn.
Lefrançois & Gandon, ULiS.
The three layersowl:Class owl:ObjectProperty xsd:boolean
owl:intersectionOf
owl:unionOf owl:hasSelf
owl:propertyChainAxiomis-aOW
Lis-a
range
:ILexicalUnit :ISemanticRelation
:ILexicalPrimitive
:onISemanticRelation
:isObligatorycore
-IL
exiM
On
lay
er
:allValuesFromdomain
range
is-a
subClassOf
subClassOf subClassOf
range
domain
domain
ILex
icO
n l
ayer
:hasEntity
:Person
:Entity
:Alive
trueonISemanticRelationallValuesFrom
intersectionOf
allValuesFrom
isObligatory
is-a
onISemanticRelation:State
is-a
Dat
a-la
yer
:Mary01 :Alive01hasEntity
MSW 2011 - 33
B
CA
A is the intersection of B and C
A B
A is a subClass of B
A B
A is linked to B through property p
p
property
A B
A is an instance of B
class/instance
MSW 2011Lefrançois & Gandon, ULiS. - 35
Die Die:Die01
Person:Mary01
IRs RDFILexicOn
1.Person
ILUc ILUi
MSW 2011Lefrançois & Gandon, ULiS. - 36
Semantic
Acta
nts
Rely on
non-INTERLINGUAL features
Conceptual participants
MSW 2011Lefrançois & Gandon, ULiS. - 37
Die< EventEnd<
1.Time
MSW 2011Lefrançois & Gandon, ULiS. - 38
Die< EventEnd<
ConP inheritance
1.Time
1.Time
MSW 2011Lefrançois & Gandon, ULiS. - 39
Die< EventEnd<
ConP inheritance
1.Time
ilexicon:Event a ileximon:ILexicalUnit,owl:intersectionOf ([ a ileximon:ILexicalPrimitive,
ileximon:onISemanticRelation ilexicon:hasTime,ileximon:allValuesFrom
ilexicon:Time,ileximon:isObligatory
"true"^^xsd:boolean.]...).
ilexicon:Die a ileximon:ILexicalUnit,owl:intersectionOf ( ilexicon:End
...).
ilexicon:End a ileximon:ILexicalUnit,owl:intersectionOf ( ilexicon:Event
...).
MSW 2011Lefrançois & Gandon, ULiS. - 40
Die End<
State1.Alive<1.
1.Person Entity
1.Time
1.Person< 1.
ConP partial inheritance
ilexicon:Die a ileximon:ILexicalUnit,owl:intersectionOf ( ilexicon:End
[ a ileximon:ILexicalPrimitive,
ileximon:onISemanticRelation ilexicon:hasState,ileximon:allValuesFrom
ilexicon:Alive,ileximon:isObligatory
"true"^^xsd:boolean. ] ...).
=?
MSW 2011Lefrançois & Gandon, ULiS. - 41
Die
1.Alive
1.Person
1.Time
1.Person
? ConP composition
Named ConP slots(an infinite number of) lexicalized conceptual relations
hasDead
hasTime
hasEntity
hasState
Die End<
=?
MSW 2011Lefrançois & Gandon, ULiS. - 42
Die
1.Alive
1.Person
1.Time
1.Person
hasDead
hasTime
hasEntity
hasState
hasState / hasEntity
hasDead
(hasState/hasEntity)<hasDead
is equivalent to:
ConP composition
Ilexicon:hasDead owl:propertyChainAxiom ( ilexicon:hasState, ilexicon:hasEntity),rdfs:domain ilexicon:Die,rdfs:range ilexicon:Person.
MSW 2011Lefrançois & Gandon, ULiS. - 43
Kill<Cause<Event
MSW 2011Lefrançois & Gandon, ULiS. - 44
hasEvent
1.Die<Event1.Person
hasAgent
Kill<Cause<Event
1.TimehasTime
MSW 2011Lefrançois & Gandon, ULiS. - 45
hasEvent
1.Die<Event1.Person
hasAgent
Kill
1.Person
hasDead
hasTime
1.Time
1.TimehasTime
MSW 2011Lefrançois & Gandon, ULiS. - 46
1.PersonhasAgenthasEvent
1.Die
ConP composition
1.Time
1.Person
hasDead
hasKilled
Kill
hasTime
1.Time
( hasEvent / hasDead ) < hasKilled
hasTime
MSW 2011Lefrançois & Gandon, ULiS. - 47
1.PersonhasAgenthasEvent
1.Die
1.Person
hasDead
hasKilled
Kill
hasTime
1.Time
? ConP merging
1.TimehasTime
MSW 2011Lefrançois & Gandon, ULiS. - 48
1.PersonhasAgenthasEvent
1.Die
1.Person
hasDead
hasKilled
Kill
hasTime
1.Time hasKillTime
ConP merging
hasTime1.Time
( hasEvent / hasTime ) < hasKillTime
MSW 2011Lefrançois & Gandon, ULiS. - 49
1.PersonhasAgenthasEvent
1.Die
hasTime1.Time
1.Person
hasDead
hasKilled
KillhasTime
hasKillTime
ConP merging( hasEvent / hasTime ) < hasKillTime < hasTime
MSW 2011Lefrançois & Gandon, ULiS. - 50
1.PersonhasAgenthasEvent
1.Die
hasTime1.Time
1.Person
hasDead
hasKilled
KillhasTime
hasKillTime
Optional / obligatory ConP
hasBeneficiary
?.Person
MSW 2011Lefrançois & Gandon, ULiS. - 51
Infanticide<Kill
MSW 2011Lefrançois & Gandon, ULiS. - 52
Relation between two ConPs(hasRelation/hasEntity) < hasKilledChild < hasKilled(hasRelation/hasObject) < hasKillerParent < hasAgent
1.PersonhasAgent
1.Person1.Parent<Relation<State
hasRelation
hasEntity
hasObject
Infanticide<KillhasKillerParent
hasKilled
hasKilledChild
MSW 2011Lefrançois & Gandon, ULiS. - 53
Inference in the data : (IRs RDF)
Person:John01hasAgent
Person:Mary01
hasKilled
Parent:p01
hasRelation
hasEntity
hasObject
Infanticide:i01hasKillerParent
hasKilledChild
hasEvent
Die:d01
hasDead
hasTimeTime:t01
hasTime
hasKillTime
ConP inheritance
ConP partial inheritance
ConP composition
ConP merging
Optional / obligatory ConP
Relation between two ConPs
Named ConP slots
Formal definitions of ILUCs
Conceptual participants
(an infinite number of) lexicalized conceptual relations
MSW 2011Lefrançois & Gandon, ULiS.
Lefrançois & Gandon, ULiS.
Three layersowl:Class owl:ObjectProperty xsd:boolean
owl:intersectionOf
owl:unionOf owl:hasSelf
owl:propertyChainAxiomis-aOW
Lis-a
range
:ILexicalUnit :ISemanticRelation
:ILexicalPrimitive
:onISemanticRelation
:isObligatorycore
-IL
exiM
On
lay
er
:allValuesFromdomain
range
is-a
subClassOf
subClassOf subClassOf
range
domain
domain
ILex
icO
n l
ayer
:hasEntity
:Person
:Entity
:Alive
trueonISemanticRelationallValuesFrom
intersectionOf
allValuesFrom
isObligatory
is-a
onISemanticRelation:State
is-a
Dat
a-la
yer
:Mary01 :Alive01hasEntity
MSW 2011 - 55
B
CA
A is the intersection of B and C
A B
A is a subClass of B
A B
A is linked to B through property p
p
property
A B
A is an instance of B
class/instance
ULiS: A Universal Linguistic System - to redesign Pivot-based NLP techniques - 100% using the Semantic Web Formalisms - compliant with the Meaning-Text Theory
ILexicOn: The Interlingual Lexical Ontology - formal lexicographic definitions
- Lexical functions- Populate: SUMO, FrameNet…- SLexicOn- Rules, SPIN & VISL-CG3
Thank YouMaxime Lefrançois & Fabien Gandon
Edelweiss – INRIA Sophia-Antipolis – France{Maxime.Lefrancois|Fabien.Gandon}@inria.fr
This presentation onhttp://maxime-lefrancois.info