From: Haitao Liu <htliuyeah.net>
Subject: Selected Lexical and Grammatical Issues in the Meaning-Text Theory
E-mail this message to a friend
Announced at http://linguistlist.org/issues/18/18-1053.html
EDITOR: Wanner, Leo TITLE: Selected Lexical and Grammatical Issues in the Meaning-Text Theory. SUBTITLE: In honour of Igor Mel'cuk PUBLISHER: John Benjamins Publishing Company YEAR: 2007
LIU Haitao, Institute of Apllied Linguistics, Communication University of China.
SUMMARY The Meaning-Text Theory (MTT) is a linguistics framework based on the idea that language is a means to express meaning by texts. MTT is a lexicon-centered and dependency-based theory incorporating semantics, syntax, morphology and lexis. Igor Mel'cuk is one of the founders of MTT. This volume, prepared on the occasion of Igor Mel'cuk's 70th birthday, tries to offer a cross-section of the current advances in MTT and its applications.
The volume includes 10 articles. They have a common leitmotif, which is pointed out by Leo Wanner (editor) in the preface: ''What is the nature of the units in a linguistic model; how are these units organized in linguistic structures, and how are these structures to be processed''(xv). Obviously, these questions should be addressed by all models with the attributive ''linguistic''. In this volume, the authors try to find the answers under the MTT.
Following the Preface of Leo Wanner, David Beck analyzes in detail different types of ideophones in Upper Necaxa Tonac for making a reasonable lexical classification. It seems that this article is not as directly relevant to MTT as other articles in the volume; however, the content is interesting, because ideophones are often ignored in mainstream linguistics.
The notion ''lexical function'' (LF) possibly is one of the most innovative ideas in MTT. Wanner has also contributed an edited volume dedicated to LF in the same linguistic series (Studies in Language Companion Series) at the same publisher (Wanner 1996). In this volume, there are six articles dealing with different aspects of LFs, which rightly emphasize LF's core role in MTT.
Alain Polguere focuses on the issue of lexical function standardness. Standard LFs are important not only due to their standardized name, but also they are the component of the formal model with the most universal value. Polguere proposes a revised definition for the notion of standard lexical function and put the emphasis on one tentative candidate to standardization: De nouveau 'again' for giving insight on how lexical functions should be dealt with as both a theoretical and practical descriptive tool and offering data on lexical relations in French.
The syntactic representation of Support Verb Constructions (SVCs) is a challenge for any syntactic theory. Margarita Alonso Ramos investigates the possibilities of the distribution of syntactic actants (SyntAs) and their correspondence with semantic actants (SemAs) in the Oper-group of LFs, which models SVCs in MTT. The results show that at the deep-syntactic level all of the noun's actants, except the first, can either remain with the keyword or be attached to the LF-lexeme itself.
Following the work of Mel'cuk and Wanner (1996) on the correlation between the semantics of emotion nouns and the collocations they occur in and the inheritance mechanisms that take this correlation into account, Francis Grossmann and Agnes Tutin examine the regularity of the LF Magn when applied to French nouns of 'joy'. They argue that this correlation can be made explicit such that the collocability between two lexemes is at least partially predictable.
Marie-Claude L'Homme proposes to apply models within Explanatory and Combinational Lexicology (ECL) to analyze and describe terms in terminology. She demonstrates how the definitional zone, the valency zone, and the lexical co-occurrence zone from the ECL should look in a terminological unit.
MTT is a linguistic theory born from early projects of Machine Translation (MT) in the USSR (Mel'cuk 2000). Thus, it is interesting to know the current state of MTT and LF applied to Natural Language Processing (NLP) in Russia and relevant to Russian. The article by Jury D. Apresjan, Igor M. Boguslavsky, Leonid L. Iomdin and Leonid L. Tsinman provides readers the necessary information on the use of LFs in fully functional advanced NLP-applications, including parsers, high quality machine translation, and a system of paraphrasing and computer-aided learning of lexica. According to the authors, LFs can be used as a disambiguation tool in parsing, and a theoretical fundamental for building the systems of paraphrasing utterances and of learning lexica; it is also helpful to finding idiomatic equivalents in MT. ''In this way, this work is contributing to converting linguistics into an experimental science''(230).
Collocation is a serious problem for learning and understanding a language. For computer understanding of a text, we should also teach the machine knowledge of collocations. Leo Wanner and Bernd Bohnet propose a method that identifies collocations as instances of a specific LF or classify the collocations based on the typology of LFs. The experiments with three machine learning techniques show that these techniques are able to provide material for LF-oriented collocation lexica which are very useful for many NLP applications. Compared with other pure statistic criteria, LF's method, that is semantic-based, has two advantages: ''1. it is not bound to the frequency with which a candidate bigram occurs in the corpus; 2. it naturally generalizes over collocates with the same meaning''(260).
Jasmina Milicevic deals with the problem of the production of paraphrases (or synonymous sentences) within MTT. She proposes a new type of paraphrasing rules operating at the level of semantic representations of sentences. These semantic paraphrasing rules can perform the tasks which can not be captured at the syntactic level. Although she can not strictly prove that semantic equivalence rules are necessary, her examples suggest the usefulness of these rules.
Kim Gerdes and Sylvain Kahane investigate the notion of phrase in non-phrase structure grammars. It is interesting not only because there is no phrase notion in dependency grammars traditionally, but also because it is useful to link word order with syntactic representation. Following Tesniere and Mel'cuk, the authors defend the idea that word order must be separated from the syntactic representation proper and that phrases are only relevant with word order. This article introduces a new notion ''topological phrase'' and distinguishes it from the classical notion of phrase in X-bar syntax. In contrast to Chomskyan models, the proposed ''phrase'' is strictly separated from the syntactic structure, although it plays the important role of bearing the information on word order. The authors argue that the phrase must be considered as a distinct linguistic unit at the deep-morphological level of representation in the MTT-model. The article is an important contribution to MTT as a comprehensive linguistic theory. It is also helpful to other linguistic theories based on dependency principle.
In the framework of MTT, ''parsing'' means the transition between the Deep-Morphological Structure (DMorphS) and Surface-Syntactic Structure (SSyntS), which is also called the ''equative approach''. It is a complex procedure, because mapping a DMorphS to an equivalent SSyntS has to use the following information: surface-syntactic well-formedness criteria, patterns for elementary phrases, word order restrictions, government patterns of lexeme, lexical functions, etc. Alexis Nasr tries to offer a more efficient automatic realization of the DMorphS->SSyntS transition. He proposes a grouping of the different knowledge sources necessary to perform such transition. Based on the complex object ''elementary tree'' and ''attachment'' operation, Nasr presents a grammar defined by a set of elementary trees and the attachment operation and shows how this generative grammar is used for parsing. This approach is influenced by the Tree Adjoining Grammar (TAG, Joshi and Schabes 1997), although Nasr's operation seems simpler than of TAG.
EVALUATION In the framework of MTT, the language is presented as a multi-stage mechanism in which each module ensures the correspondence between two adjacent levels of utterance representation. Classical Meaning-Text Theory assumes the following representations: Semantic, Deep syntactic, Surface syntactic, Deep morphological, Surface morphological, Deep phonological, Surface phonological representations. A short Guide to MTT is available online (Milicevic, 2006).
This volume is very useful to know the current state of Meaning-Text Theory. Although it does not include a complete panorama of this interesting linguistic theory, it reflects two hot-spots of MTT: lexical function and its application to NLP.
The volume is useful to all workers in MTT, it is also helpful to NLP lexicographers and researchers, in particular those who are searching for theoretical fundamentals of linguistic practice. Readers unfamiliar with MTT in general and LFs in particular are advised to first read Mel'cuk (1988, 1997), Wanner (1996) and Milicevic(2006).
The volume is edited and published in almost perfect quality, although there are a few bugs in the preface. For example, the editor writes ''the second part of the volume, which consists of four articles... The first article of this part...'' (xvii). This should read: ''the second part of the volume, which consists of three articles... The second article of this part...''
In summary, the volume provides us an updated MTT, one of the most compelling linguistic theories for human languages.
REFERENCES Joshi, Aravind K. and Yves Schabes. (1997) Tree-Adjoining Grammars. In _The Handbook of Formal Languages_, volume 3, Springer-Verlag. pp. 69-123.
Mel'cuk, Igor. (1988) _Dependency Syntax: Theory and Practice_. Albany: State University of New York Press.
Mel'cuk, Igor. (1997) Vers une linguistique Sens-Texte. Lecon inaugurale. Paris: College de France. http://www.olst.umontreal.ca/FrEng/melcukColldeFr.pdf
Mel'cuk, Igo.r (2000) Machine Translation and Formal Linguistics in USSR. In John Hutchins (ed.) _Early Years in Machine Translation_. John Benjamins Publishing Company. pp. 205-226.
Mel'cuk, Igor and Leo Wanner. (1996) Lexical Functions and Lexical Inheritance for Emotion Lexeme in German. In Leo Wanner (ed.) _Lexical Functions in Lexicography and Natural Language Processing_. Amsterdam/Philadelphia: John Benjamins Publishing Company. pp. 209-278.
Milicevic, Jasmina. (2006) A short Guide to the Meaning-Text Linguistic Theory. _Journal of Koralex_, Vol. 8, 187-233.
Wanner, Leo. (1996, ed.) _Lexical Functions in Lexicography and Natural Language Processing_. Amsterdam/Philadelphia: John Benjamins Publishing Company.
ABOUT THE REVIEWER LIU Haitao is professor of applied and computational linguistics at the Communication University of China (CUC). His research interests include computational linguistics, syntactic theory and language planning.