Linguistic Form and its Computation

The aim of this book is to present results of the Collaborative Research Center 'Linguistic Foundations for Computational Linguistics' at the universities of Stuttgart and Tübingen. The goal of the Center has been to foster interaction between theoretical and computational linguistics. Its point of departure was the idea that on the one hand, computational linguistics and its applications should be based on a theoretically sound analysis of natural language; and that, on the other, the work in natural language processing can and should contribute to the development of theoretical linguistics.

The papers gathered in the volume cover the following areas: Syntax, Syntax-Semantics Interface, Syntax-Pragmatics Interface, Discourse, Formal Properties of Grammars and Probabilistic Methods for Lexicon Induction. All the papers endeavor to present their results in a form exact enough that they could in principle be implemented. Only a few papers, however, are based on actual implementations.

An important role plays the idea that linguistic description involves not just one level but several at once and that each level comes with its own representations. The interfaces between the levels are as much an integral part of the theory as the representations between which they mediate.

The relationship between syntax and semantics in this volume is explored mainly within the context of Discourse Representation Theory (DRT), because this semantic framework has been the formal conceptual background for a good part of the work in semantics and pragmatics within the Center. One of the central tenets of DRT is that the 'unit of analysis' for a theory of meaning is the discourse rather than the single sentence.

One of the most serious challenges for natural language processing is ambiguity. Parsers for grammars with large coverage often produce a huge number of syntactic analyses for one sentence. In this volume the problem of ambiguity is examined from an optimality-theoretic point of view and from a probabilistic one. The authors show the close relation between optimality-theoretic disambiguation and probabilistic language models. In addition the use of partial, or underspecified representations in syntax and semantics is explored.

Classical complexity questions remain important for practical as well as theoretical reasons. Two papers investigate the complexity of syntactic theories belonging to the Government and Binding family.

Christian Rohrer and Hans Kamp are professors at the Institute for Computational Linguistics (IMS) of the University of Stuttgart. Antje Roßdeutscher is a senior researcher at the IMS.

Contents

12/1/2001