Centre for Computational Linguistics: Projects
LINGUADUCT: Contextual interpretation of natural language texts by means of abductive reasoning and
inductively acquired knowledge
Time Span: 1997 - 2000
FWO-Vlaanderen & Nationale Loterij
F. Van Eynde, J. Riche & N. Schelkens
Other participants:
Dept. Computerwetenschappen (K.U.Leuven), Centrum voor Nederlandse Taal en Spraak (U.I.Antwerpen), Vakgroep Informatica (U.I.Antwerpen)
Aim
The formal specification of a model for the contextual interpretation of
natural language texts which makes use of the latest techniques in
abductive reasoning and inductive acquisition of linguistic knowledge.
Objectives
A major challenge in the modelling of text understanding is that texts
underdetermine their interpretation.
We will concentrate on the use of inferencing techniques for the
resolution of ambiguities and for the retrieval of cross-clausal
relations, esp. temporal ones. To this end we will
- Develop a model for the contextual interpretation of temporal
expressions. This model has to facilitate the retrieval of
cross-clausal temporal relations in texts.
- Develop a model for the automatic extraction of linguistic knowledge
from large corpora. This knowledge has to facilitate the
contextual disambiguation of words and phrases.
- Develop techniques for reasoning with underspecified representations
and for augmenting them by making use of abduction.
- Develop techniques for the induction of knowledge from large
data bases.
Design and methodology
- Merging a Typed Feature Structure linguistic formalism, such as
Head-driven Phrase Structure Grammar, with a discourse oriented
framework, like Discourse Representation Theory, and use the
resulting representation language for (1) the morphological,
syntactic and semantic analysis of temporal expressions, and (2)
the application of reasoning procedures on the representations
- Application of data mining methods, such as memory-based learning and
rule induction, to large annotated text corpora for the extraction
of the kind of linguistic knowldege that is needed for contextual
disambiguation. Special attention will be paid to the modelling of
sub-regularities and exceptions (cf. the use of probabilistic
techniques)
- Formal specification and implementation of a temporal logic (Open
Event Calculus) and of techniques for abductive reasoning in the
Logic Programming paradigm; application to
the contextual interpretation of temporal expressions
- Merging methods of inductive logic programming with the modelling of
relational graph-based databanks, and explore the potential of this
combination for the induction of linguistic knowledge from large
text corpora.
CCL
Layout:
webmaster@ccl.kuleuven.ac.be
Information Provider: Centrum voor Computerlinguïstiek
Comments to the Author:
Frank.VanEynde@ccl.kuleuven.ac.be
(C) Copyright 1996, CCL.
All Rights Reserved.