Skip to main content

Dynamic Interpretation of Natural Language

Objective

The DYANA Action aimed to answer the following questions:
-What declarative theory of grammar (extending from phonetics to discourse) would best support incremental interpretation and top-down information flow?
-How can existing unification-based grammar formalisms be developed so as to increase their ability to express high-level generalisations, while retaining a clear semantics and computational interpretation?
-How is the informational content of an utterance integrated into the current discourse context?
-What formal models can adequately capture the defeasibility and non-monotonic character of human reasoning?
-To what extent is it possible to synthesise the formal methods used for modelling partial information in speech, grammar, semantics and reasoning?
Fundamental questions about the integration of logic, natural language and speech were addressed, focussing on the themes of partial information and dynamic interpretation in natural language processing, with particular attention to developing a computational and cognitively motivated model of how spoken language is understood.

The available deliverables cover the following areas:
grammar development (speech and prosody involving parametric variation in German and Romance, flexible montague grammar, psycholinguistic studies in the incremental interpretation of speech and the labelling of intonation contours);
meaning, discourse and reasoning (generalized quantifier logic for naked infinitives, dynamic predicate logic, dynamic montague grammar, the semantics of progressives and perfects, conditionals, and defaults and belief revision);
logic and computation (a comprehensive unification formalizm for linguistic applications constraint logic programming for natural language analysis and foundations of unification based formalizms);
a prototype intonational parser and a (nearly) deterministic parser.
APPROACH AND METHODS
The work was divided into three interdependent themes:
-Grammar development, speech and prosody, emphasising the development of formal models, the abstract specification and parametric variation of linguistic structure, the incorporation of intonational structure into the representation of utterances, and th e investigation of the interplay of intelligibility, predictability and discourse context in speech recognition and production.
-Meaning, discourse and reasoning, addressing the further development of the dynamic and partial models of meaning that have been emerging in the past decade. In addition, non-monotonic logic will be used, taking into account conditional reasoning and in feasibility in lexical concept combination. The Action aimed to formalise default interpretation, which plays a role in various aspects of linguistic structures.
-Logic and computation, directed towards the development of formal mathematical and computationally tractable models of natural language, encompassing recent extensions of unification formalisms.
In the final phase of this Action, work on classes of algebraic description spaces and a comprehensive unification formalism for linguistic applications was to be incorporated and implemented, as well as a proof theory for temporal formalisms.
PROGRESS AND RESULTS
The available deliverables cover the following areas:
-Grammar development. speech and prosody: parametric variation in German and Romance; Flexible Montague Grammar; psycholinguistic studies in the incremental interpretation of speech; the labelling of intonation contours.
-Meaning, discourse and reasoning: generalised quantifier logic for naked infinitives; dynamic predicate logic; Dynamic Montague Grammar; the semantics of progressives and perfects, conditionals, defaults and belief revision.
-Logic and computation: a comprehensive unification formalism for linguistic applications; constraint logic programming for natural language analysis; foundations of unification-based formalisms.
Other deliverables include a prototype intonational parser, and a (nearly) deterministic parser.
Some of this work is being continued in DYANA-2, ESPRIT Basic Research Project 6852.
POTENTIAL
The research themes that constitute the DYANA Action aim to meet some of the challenges that lie ahead on our way to computer systems that use spoken or written human language. Although the major aim of the Action is to carry out fundamental research in natural language processing, work is also directed towards the development of prototypes and proof-of-concept implementations.

Coordinator

University of Edinburgh
Address
2 Buccleuch Place
EH8 9LW Edinburgh
United Kingdom

Participants (4)

Eberhard-Karls-Universität Tübingen
Germany
Address
Biesinger Straße 10
72070 Tübingen
Technische Universitaet Muenchen
Germany
Address
Arcisstrasse 21
80333 Muenchen
UNIVERSITEIT VAN AMSTERDAM
Netherlands
Address
Kloveniersburgwal, 72
1012 CZ Amsterdam
Universitaet Stuttgart
Germany
Address
Keplerstrasse 7
70174 Stuttgart