Semantics and discourse

Directorate-General For Translation European Commission

Description

An NLP perspective
Information technology and telecommunications
Information - Documentation

Sujets

• Information processing
• Linguistics
• Machine translation
• Programming language
• Terminology

Informations

• Publié par: EUROPEAN-COMMISSION
• Nombre de lectures: 5
• Langue: English
• Poids de l'ouvrage: 3 Mo

Extrait

STUDIES IN MACHINE TRANSLATION
AND NATURAL LANGUAGE PROCESSING
Volume 9
SEMANTICS AND DISCOURSE:
A NLP PERSPECTIVE
Edited by
Frank VAN EYNDE
and
Valerio ALLEGRANZA
European Commission Studies in machine translation
and natural language processing
Published by:
Office for Official Publications
of the European Communities Managing editor
Erwin Valentini (CEC), Luxembourg
Editorial board
Doug Arnold
(Department of Language and Linguistics, United Kingdom)
Nicoletta Calzolari
(Istituto di Linguistica Computazionale, Italia)
Frank Van Eynde
(Nationaal Fonds voor Wetenschappelijk Onderzoek, België)
Steven K rauwer
(Rijksuniversiteit Utrecht, Nederland)
Bente Maegaard
(Center for Sprogteknologi, Danmark)
Paul Schmidt
(Institut für Angewandte Informationsforschung, Deutschland)
Luxembourg: Office for Official Publications of the European Communities, 1995
ISSN 1017-6568
© ECSC-EC-EAEC, Brussels · Luxembourg, 1995
Printed in Italy Volume 9
Semantics and Discourse:
an NLP Perspective
Edited by
Frank Van Eynde and Valerio Allegranza
European Commission Volume 9
Semantics and Discourse:
an NLP Perspective
Editors
Frank Van Eynde
Valerio Allegranza
Contents
FRANK VAN EYNDE
Introduction 7
VALERIO ALLEGRANZA
Formal Semantics for Natural Language: a Non-Technical Overview 11
STEPHEN THOMAS
Speech Act Theory and its Relevance for Computational Linguistics 44 VALERIO ALLEGRANZA
Discourse Representation Theories 56
HELEN DE HOOP
Dynamic Discourse Semantics 90
PETER I. BLOK
From Structured Meanings to Formal Pragmatics 109
JACQUES DURAND
Systemic Functional Linguistics and Discourse: a Critical Appraisal 143
STEPHEN THOMAS
Rhetorical Structure Theory 159
JOHAN Bos
Anaphora Resolution and Discourse Segmentation in Artificial Intelligence 175 FRANK VAN EYNDE
Introduction
It can be said, without exaggeration, that the future of Natural Language Processing (NLP)
is in the processing of discourse. At present, though, the large majority of NLP systems
concern the analysis, generation or translation of individual sentences. This limitation
is defensible on grounds of feasibility, since the problems with sentence-based NLP are
sufficiently complex for their own sake. In the long run, however, this limitation is not
tenable, since most interesting applications require the processing of texts, rather than of
isolated" sentences.
In order to see where the field of discourse processing stands today (1995), it is useful to
make a comparison with the field of computational syntax. Surveying the developments in
the latter one can distinguish three phases.
In the first phase, the processing of sentences was mainly seen as an engineering exercise, in
which linguistic considerations were by and large ignored; the first generation of Machine
Translation systems, for instance, is characterised by this a-linguistic attitude.
The second phase is characterized by a rapprochement between linguistics and NLP, con­
ditioned by the fact that syntax had taken a formal turn (cf. the Chomskyan revolution) and
that computer science had developed programming languages which are more appropriate
for linguistic analysis than the machine orienteds of the early days; the second
generation of Machine Translation systems with their monolingual analysis and synthesis
modules are typical of this phase.
The third phase is characterised by a blurring of the distinction between the formalisms used
for linguistic description and the formalisms used for the processing of linguistic data. With
the emergence of the unification based approach in NLP (cf. Definite Clause Grammar,
Functional Unification Grammar, PATR) and the feature based approach in linguistics (cf.
Lexical Functional, Generalised Phrase Structure Grammar, Head-driven Phrase
Structure Grammar) the gap between NLP and theoretical linguistics has become so small,
that the distinction between formal and computational syntax has lost much of its relevance.
As for the field of discourse processing, we are somewhere at the beginning of phase
two. From the side of the theory, we are witnessing the emergence of discourse models Studies in MT and NLP, Volume 9
which are sufficiently formalised to be of some use for NLP, but -as they stand- their
coverage is rather restricted and their commitment to computation is often limited to
such general statements as "it is compositional, hence it will be implementable". At the
computational side, there is a growing number of applications which can be seen as more or
less faithful implementations of existing theories, but the majority of systems concentrate
on the development of heuristics forsolving specific problems, such as anaphora resolution,
and lack the generality which is required of a theory.
In short, we are still at the beginning of a rapprochement between theory and application.
However, to the extent that the developments in computational syntax can serve as a guide,
it is to be expected that the major innovations and breakthroughs in discourse processing
will come from a further integration of theoretical models and task-specific applications.
In conformity with this expectation the present volume focuses on those theories of dis­
course which have already been used for implementation and on those applications which
aim for generality. Reflecting the state of the art in the field, the emphasis will be on seman­
tic aspects of discourse processing, but syntactic and pragmatic aspects will be discussed
as well.
What this volume provides, is not so much a contribution to any particular theory nor
a description of an implemented system, but rather a survey of existing theories and a
succinct assessment of their relevance for NLP.
Historically, the study of discourse starts with Classical Rhetorics, but for centuries this
approach has been prescriptivist, focusing on such issues as how to construct a convincing
argument or how to manipulate an audience. To the extent that these prescriptions are
based on the analysis of existing models, they also have a descriptivist vein, and current
Argumentation Theory, which can be seen as the successor of Classical Rhetorics, is
keen on stressing this. But, even with the change of emphasis from text production to text
interpretation, and from prescription to description, the rhetorical approach lacks the degree
of explicitness and generality which would make it useful as a model for computational
discourse processing. Being primarily concerned with interpretation by human agents, it
assumes and presupposes a lot about the more basic aspects of text interpretation, which
one cannot take for granted when developing a formal theory of discourse. For this reason
we have not included a separate chapter on the rhetorical approach (a brief discussion can
be found, though, in the chapter on Rhetorical Structure Theory).
More relevant for the field of discourse processing is the line of research which is known as
Formal Semantics. Its origins can be traced back to the work of logicians like Frege, Russell
and Tarski, and its first systematic application to natural languages is due to Montague. In
"Formal semantics for natural language: a non-technical oveì-view " Valerio Allegranza
sketches a broad survey of this line of research, paying special attention to the context
dependent aspects of interpretation and to the integration of syntax and semantics.