A SHORT GUIDETO THE MEANING-TEXT LINGUISTIC THEORY

JASMINA MILIĆEVIĆ DALHOUSIE UNIVERSITY - HALIFAX (CANADA)

2006, Journal of Koralex, vol. 8: 187-233

Contents

0. Introduction (1-2)1. Postulates and methodological principle (2-4)2. Meaning-Text models (4-6)3. Illustration of the linguistic synthesis in the

Meaning-Text framework (6-27)4. Summary of MTT’s main features (27-30)5. Basic Meaning-Text bibliography (30-36)

0. Introduction

• MTT = theoretical framework for the construction of models of languages

• Launched in Moscow (Žolkovskij & Mel’čuk 1967)

• Developed in Russia, Canada, Europe• Formal character computer applications• Relatively marginal

1. Postulate 1

• “Natural language is (considered as) a many-to-many correspondence between an infinite denumerable set of meanings and an infinite denumerable set of texts.” (2)

{SemRi} <=language=> {PhonRj} │0 < i, j ∞

Postulate 2

• “The Meaning-Text correspondence is described by a formal device which simulates the linguistic activity of the native speaker—a Meaning-Text Model.”(3)

Postulate 3

• “Given the complexity of the Meaning-Text correspondence, intermediate levels of (utterance) representation have to be distinguished: more specifically, a Syntactic and a Morphological level.”(3)

Methodological principle

• “The Meaning-Text correspondence should be described in the direction of synthesis, i.e., from Meaning to Text (rather than in that of analysis, i.e., from Text to Meaning).” (3)

WHY?

1. Producing speech is an activity that is more linguistic than understanding speech;

2. Some linguistic phenomena can be discovered only from the viewpoint of synthesis (ex: lexical co-occurrence = collocations).

• Corollary:• study of paraphrases (and lexicon) occupies a central place

in the M-T framework.

Paraphrase

• Synonymy = fundamental semantic relation in natural language “to model a language means to describe its synonymic means and the ways it puts them in use”.

• Meaning = invariant of paraphrases • Text = “virtual paraphrasing” • Lexical paraphrase semantic decomposition

of lexical meanings

Semantic decomposition of ‘criticize’

• (definiendum): ‘X criticizes Y for Z’ • ≈ (definiens):

– ‘Y having done21 Z which X considers2 bad2 for Y or other

people1,– and X believing3 that X has good1

1 reasons12 for

considering2 Z bad2, ||– X expresses3

1 X’s negative11 opinion1 of Y because of Z(Y),

– specifying what X considers2 bad2 about Z,– with the intention2 to cause2 that people1 (including Y) do

not do21 Z.’

2. Meaning-Text Models: Characteristics

• Equative = transductive generative (Postulate 1)

• Completely formalized (Postulate 2)• Stratificational model (Postulate 3)

MTM Architecture

(Neuvel)

Representations

Neuvel.net, (adapted from Mel'chuk 1988: 49)

2. MTM: peripheral structures

• Reflect different characerizations of the central entity = provide additional information relevant at each level.

• Peripheral: they do not exist independently of the central structure.

• Purpose: to articulate the SemS into a specific message, by specifying the way it will be ‘packaged’ for communication.

Central and peripheral S / level of R

• SemR = <SemS, Sem.CommS, RhetS, RefS>• DSyntR = < DSyntS, Dsynt-CommS, DSynt.-

ProsS, Dsynt-AnaphS)• SSyntR = <SSyntS, SSynt-CommS, SSynt-ProsS,

SSynt-AnaphS>• DMorphR = <DMorphS, Dmorph-ProsS>

2. MTM: rules

3. Illustration: Linguistic Synthesis

• Synthesis: 1 SemR (X 2) 3 PhonR (X 2)• SemR [1]: Theme = media PhonR (1 a, b, c)• SemR [2]: Theme = decision PhonR (2 a, b, c)

SemR’s central structure = SemS

• A SemS represents the propositional meaning of a set of paraphrases.

• SemS = network: nodes and arcs• Nodes: labeled with semantemes.• Arcs: labeled with numbers (predicate-

argument relations).

SemS (example)

Peripheral structure Sem-CommS

• Sem-CommS represents the communicative intent of the Speaker.

• Formally, Sem-CommS = division of the SemS into communicative areas, each marked with one of mutually exclusive values.

Eight communicative oppositions

• Thematicity = {Theme, Rheme, Specifier}• Giveness = {Given, New}• Focalization = {Focalized, Non-Focalized}• Perspective = {Backgrounded, Foregrounded, Neutral}• Emphasis = {Emphasized, Neutral}• Assertiveness = {Asserted, Presupposed}• Unitariness = {Unitary, Articulated}• Locutionality = {Communicated, Signaled, Performed}

Other peripheral Sem-structures

• Sem-RhetS represents the Speaker’s rhetorical intent.

• Sem-RefS = set of pointers from semantic configurations to the corresponding entities in the real world.

Theme: media

(1) a. [The media]T [harshly criticized the Government for its decision to increase income taxes]R

b. [The media]T [seriously criticized the Government for its decision to raise income taxes]R

c. [The media]T [leveled harsh criticism at the Government for its decision to increase income taxes]R

Theme = Media

Theme = government’s decision

(1) a. [The government’s decision to increase income taxes]T [was severely criticized by the media]R

b. [The government’s decision to raise income taxes]T [drew harsh criticism from the media]R

c. [The government’s decision to increase income taxes]T [came under harsh criticism from the media]R

Theme = government’s decision

Syntactic dependency

• Relation of strict hierarchy• Characteristics:

– Antireflexive – Antisymmetric– Antitransitive

Syntactic structure

• Tree• Nodes labeled with lexical units; not linearly

ordered• Top node does not depend on any lexical unit

in the structure, while all other units depend on it, directly or indirectly.

• Arcs (= branches) labeled with dependency relations

DSyntS

• Nodes: labeled with deep lexical units (≠ pronouns and ‘structural words’) subscripted for all meaning-bearing inflections.

• Branches: labeled with names of deep syntactic dependency relations.

• Deep lexical unit = lexeme, (full) phraseme or name of a lexical function.

Lexical functions

• LF = formal tools used to model lexical relations, i.e., restricted lexical co-occurrence (= collocations), and semantic derivation. They have different lexical expressions contingent on the keyword.

• LF corresponds to a meaning whose expression is phraseologically bound by a particular lexeme L (= argument of the LF).

Lexical functions: examples

• Magn ‘intense/very’– Magn(wind) = strong, powerful– Magn(rain(N)) = heavy, torrential // downpour

– Magn(rain(V)) = heavily, cats and dogs

• S1 ‘person/object doing L’– S1(crime) = author, perpetrator [of ART ˷ ] //

criminal– S1(kill) = killer

Lexical functions: classification

1. According to their capacity to appear in the text alongside the keywords: syntagmatic (normally do) and paradigmatic (normally do not)

2. According to their generality/universality: standard (general/universal) and non-standard (neither general nor universal)

3. According to their formal structure: simple and complex

Examples

• Magn: syntagmatic, standard, simple LF • S1: paradigmatic, standard, simple LF• A YEAR that has 366 days = leap [˷] =

non-standard LF: it only applies to one keyword (year) and has just one value (leap); not universal (not valid cross-linguistically)

• CausePredPlus: complex LF

LFs realized in (1) and (2)• Magn(criticize) = bitterly, harshly, seriously, strongly //

blast• Magn(criticism) = bitter, harsh, serious, severe, strong• CausePredPlus(taxes) = increase, raise• QSØ(criticize) = criticism• QSØ(decide) = decision

• Oper1(criticism) = level [˷ at N|N denotes a person], raise [˷

against N], voice [˷]• Oper2(criticism) = come [under ˷], draw [˷ from N], meet

[with ˷]

Deep lexical units

• Do not correspond one-to-one to the surface lexemes: in the transition towards surface syntax, some deep lexical units may get deleted or pronominalized and some surface lexemes may be added.

12 Deep-Syntactic Relations

• 6 actantial DSyntRels (I, II, III,…, VI) + 1 DSyntRel for representing direct speech (=variant of DSyntRel II)

• 2 attributive DSyntRels: ATTRrestr(ictive) and ATTRqual(ificative)

• 1 Appenditive DSyntRel (APPEND): links the Main Verb to ‘extra-structural’ sentence elements (sentential adverbs, interjections,…)

• 2 coordinative DSyntRels: COORD and QUASI-COORD

DSyntR – (1a)

DSyntR – (1b)

DSyntR – (1c)

Semantic module:correspondence rules

• Lexicalization rules• Morphologization rules• Arborization rules• Communicative rules• Prosodic rules

SemR[1] DSyntRs (1a) and (1b)

Figure 10:A lex.-funct. rule

Figure 11:Arbor. rule 1

Figure 12: Arbor. rule 2

Semantic module: equivalence rules

• = paraphrasing rules1. Semanic equivalence rules equivalence

between (fragments of) 2 SemRs2. Lexico-syntactic rules: formulated in terms of

lexical functions equivalence between (fragments of) 2 DSyntRs.

Ex.: lexical-syntactic equivalence rule

From D to SSyntR: the Deep-Syntactic module

• SSyntS: dependency tree; nodes labeled with actual lexeme; branches labeled with names of language specific surface-syntactic dependency relations.

• DSyntS ≠ SSyntS: 1. Lexically: only semantically full lexemes vs all lexemes

(including full and structural words + pronouns)2. Syntactically : only universal dependency relations vs

specific dependency relations

DSyntR / SSyntR (1a)

SSyntR (1b)

SSyntR (1c)

Deep-Syntactic module: major types of rules

1. Phrasemic rules2. Deep-Syntactic rules3. Pronominalization rules4. Ellipsis rules5. Communicative rules6. Prosodic rules

6 phrasemic rules (1 a-c)

• SSyntS (1a)– 1) Magn(CRITICIZE) <=> harshly; – 2) CausPredPlus(TAXES) <=> increase

• SSyntS (1b)– 3) Magn(CRITICIZE) <=> seriously; – 4) CausPredPlus(TAXES) <=> raise

• SSyntS (1c)– 5) Oper1(CRITICISM) <=> level; – 6) Magn(CRITICISM) <=> harsh

Constraints: examples

• (3) a. The media raised harsh criticism against the Government for its decision to impose higher taxes. / The media leveled harsh criticism at the Government for its decision to impose higher taxes.

• b. The media raised harsh criticism against the Government’s decision to impose higher taxes. vs. *The media leveled harsh criticism at the Government’s decision to impose higher taxes.

• (4) ?The media raised harsh criticism against the Government for its decision to raise taxes.

DSynt-rule 1 (1a – 1b)

DSynt-rule 2 (1a-1b)

From SSyntR to DMorphR: the Surface-Syntactic Module

• DMorphS = string of fully ordered lexemes subscripted with all inflectional values

• DMorph-ProsS = specification of semantically + syntactically induced prosodies

DMorphRs (1)• Sentence (1a)

– THE MEDIApl | HARSHLY CRITICIZEact, ind, past, 3(?)sg THE GOVERNMENTsg || FOR ITSsg DECISIONsg | TO INCREASEinf INCOMEsg TAXpl |||

• Sentence (1b)– THE MEDIApl || SERIOUSLY CRITICIZEact, ind, past, 3 (?)sg THE

GOVERNMENTsg, possessive DECISIONsg | TO RAISEinf INCOMEsg TAXpl |||

• Sentence (1c)– THE MEDIApl | LEVELact, ind, past, 3 (?)sg HARSH CRITICISMsg

AT THE GOVERNMENTsg || FOR ITS DECISIONsg |TO INCREASEinf INCOMEsg TAXpl |||

SSynt-module: major types of rules

1. Linearization rules– Local (and semi-local):(5) a. [the government’s]elementary.ph. [decision]elementary.ph.

[to increase]elementary.ph. [taxes]elementary.ph.

b. [[the Government’s decision]complex ph. [to increase taxes]complex ph. ]complex ph.

– Global 2. Morphologization rules3. Prosodization rules

Example: local linearization rule (1c)

4. Main features of the MTT

1. Globality, descriptive orientation2. Semantic bases and synthesis orientation, essential role

of the paraphrase and of communicative organization3. Strong emphasis on the lexicon4. Relational approach to language: the use of

dependencies at all levels of linguistic description5. Formal character6. Stratificational and modular organization of MTMs7. Implementability: the MTT lends itself well to computer

applications

5.7 Computational Linguistics and NLP Applications

• Apresjan Ju. et al. (2003). ETAP-3 Linguistics Processor: a Full-Fledged Implementation of the MTT. In: Kahane, S. & Nasr, A., eds. (2003), 279-288.– (1992). Lingvističeskii processor dlja složnyx informacionnyx system [A

Linguistic Processor for Complex Information Systems]. Moskva: Nauka.– (1989). Lingvističeskoe obespečenie sistemy ÈTAP-2 [Linguistic Software

for the System ETAP-2]. Moskva: Nauka.• Apresjan, Ju. & Tsinman, L. (1998). Perifrazirovanie na kompjutere

[Paraphrasing on the Computer]. Semiotika i informatika 36, 177-202.• Boguslavskij, I., Iomdin. L. & Sizov. V. (2004). Multilinguality in ETAP-3.

Reuse of Linguistic Ressources. In: Proceedings of the Conference Multilingual Linguistic Ressources. 20th International Conference on Computational Linguistics, Geneva 2004, 7-14.

5.7 Computational Linguistics and NLP Applications

• Boyer, M. & Lapalme, G. (1985). Generating Paraphrases from Meaning-Text Semantic Networks. Montreal: Université de Montréal.

• CoGenTex (1992). Bilingual Text Synthesis System for Statistics Canada Database Reports : Design of Retail Trade Statistics (RTS) Prototype. Technical Report 8. CoGenTex Inc., Montreal.

• Iordanskaja, L., Kim, M., Kittredge, R., Lavoie, B. & Polguère, A. (1992). Generation of Extended Bilingual Statistical Reports. In: COLING-92, Nantes, 1019-1022.

• Iordanskaja, L., Kim, M. & Polguère, A. (1996). Some Procedural Problems in the Implementation of Lexical Functions for Text Generation. In: Wanner, L., ed., (1996), 279-297.

5.7 Computational Linguistics and NLP Applications

• Iordanskaja, L., Kittredge, R. & Polguère, A. (1991). Lexical Selection and Paraphrase in a Meaning-Text Generation Model. In: Paris, C. L., Swartout, W. R. & Mann, W. C., eds., Natural Language Generation in Artificial Intelligence and Computational Linguistics. Boston: Kluwer, 293-312.

• Iordanskaja, L. & Polguère, A. (1988). Semantic Processing for Text Generation. In: Proceedings of the First International Computer Science Conference-88, Hong Kong, 19-21 December 1988, 310-318.

• Kahane, S. & Mel’čuk, I. (1999). Synthèse des phrases à extraction en français contemporain (Du graphe sémantique à l’arbre de dépendance). T.A.L., 40:2, 25-85.

• Kittredge, R. (2002). Paraphrasing for Condensation in Journal Abstracting. Journal of Biomedical Informatics 35: 4, 265-277.

Bibliography

• MILIĆEVIĆ, Jasmina (2006): « A Short Guide to the Meaning-Text Linguistic Theory », Journal of Koralex, vol.8: 187-233.

• NEUVEL, Sylvain: Linguistic Theories> Meaning-Text Linguistics > Introduction <http://www.neuvel.net/meaningtext.htm> (8/5/2011)