IRIDIA [ Main | Projects | People | Meeting | Library ] IRIDIA

Phrase Structure Grammars for Natural Languages

Intro

Using natural languages is a crucial human ability that has so far proved difficult to transfer to computers. Crucial to our progress in this domain is a better understanding of the empirical properties of natural languages and the development of linguistic formalisms appropriate both for analyzing these properties and for computer application, such as the newly developed Head-driven Phrase Structure Grammar framework.


Natural Language Processing (NLP) is one of the central domains of investigation in artificial intelligence. Developments in this domain crucially depend on progress in our theoretical and empirical understanding of the way natural languages work. Head-driven Phrase Structure Grammar (HPSG, cf. Pollard and Sag 1987, 1994) provides a framework which is especially suitable both for purely linguistic research and for NLP applications of the results obtained. On the one hand, it rests on a careful and coherent formalism (the logic of typed feature structures) which is well-suited for computational applications, providing reversibility, declarativeness and the use of partial descriptions. On the other hand, it integrates many essential empirical and theoretical advances of the major frameworks that have preceded it. Apart from English, which is the most completely described, the HPSG framework has been adopted in studies of various other languages, e.g. German, Korean, Italian, Japanese. Among other projects, it is the framework adopted for Verbmobil a German/English/Japanese interpretation project funded by the German government.


A Phrase Structure Grammar of French

Numerous syntactic phenomena in French have been studied in depth by traditional grammarians as well as by modern generative linguists. However, neither research tradition has the resources necessary to provide a coherent and explicit model for studying the entire body of French syntactic phenomena. In collaboration with Anne AbeillÈ and DaniËle Godard (UniversitÈ Paris 7) and Ivan Sag (Stanford University), we have decided to systematically take up the principle syntactic phenomena of the French language once again, in order to obtain a detailed and reasonably formalized description which constitutes what can be termed a phrase structure grammar of French. Major results have already been obtained in the treatment of auxiliaries, causative and perception verbs, clitic complements, and extraction out of noun phrases. Our current research topics include: (i) verbal constructions: passives, reflexives (inherent, middle and true reflexives), impersonal constructions, inverted constructions, support verbs; (ii) noun phrases: argument structure, determiners, prepositions; (iii) verb complementation: infinitives and complements with que, raising vs. control verbs; (iii) negation; (iv) coordination.


English Verbal Complementation

The analysis of the types of complement structures that different verbs allow is a central topic in linguistics, involving syntactic, semantic and even morphophonological properties of verbs. For instance, though both They sent a message to Kim and They sent a message to Chicago are acceptable sentences, the acceptable variant They sent Kim a message contrasts with the unacceptable *They sent Chicago a message. Similarly the verb begin allows both infinitival and gerund complements (Kim began to eat; Kim began eating) whereas finish allows only gerunds (Kim finished eating; *Kim finished to eat). It is theoretically and practically crucial to understand to what extent such alternations are arbitrary, or can be motivated in terms of meaning or other properties of the verb. HPSG provides an ideally suited framework for discussing such alternations, since the feature structures describing verbs provide simultaneous access to syntactic, semantic and morphophonological information, thereby making it easy to state principled explanations of relevant correlations. The present project aims to explain complementation alternations in small groups of semantically similar verbs, on the basis of large corpora of examples of actual uses of the verbs obtained from large English text data banks.


Strong Generative Capacity

Crucial to a well-conceived implementation of linguistic formalisms is an understanding of the implicit semantics that underlies them. Building on Chomsky's original definition of Strong Generative Capacity (SGC) as the set of derivations produced by a grammar, we propose a framework where SGC is understood as providing model-theroretic semantic interpretations for the formalisms employed by linguistic theories and are currently applying it to a broad range of such theories.


[ Philip H. Miller ]

Selected references

P. H. Miller.
Clitics and Constituents in Phrase Structure Grammar.
New York: Garland, 1992.
P. H. Miller.
Strong Generative Capacity
The Semantics of Linguistic Formalism. Stanford: CSLI Press, to appear.
P. H. Miller et I. A. Sag.
Une analyse lexicaliste des affixes pronominaux du franÁais.
Revue quÈbÈcoise de linguistique, 1995.
A. AbeillÈ, D. Godard, and P.H. Miller.
The syntactic structure of French causative constructions.
Unpublished ms. UniversitÈ Paris 7 et UniversitÈ Libre de Bruxelles.

IRIDIA [ Main | Projects | People | Meeting | Library ] IRIDIA

Last updated Jan 15, 1996 | Comments to webmaster