Property Grammars: a new framework for non-generative approaches

Abstract:

Modern linguistic theories, formalisms and techniques have to take into consideration the fact that linguistic information is spread, incomplete and partial. This requires a flexible approach capable of explaining how different domains of linguistic analysis (prosody, pragmatics, syntax, etc.) can interact in order to make an utterance understandable. Moreover, linguistic realizations can have many different forms, more or less canonical (literary texts, SMS, spoken language, ...). It is then necessary to propose a robust and flexible solution. Usually, this question is addressed in terms of relations between structures built for each domain. However, several works (e.g. model-theoretic syntax or construction grammars) have underlined the theoretical and practical limits of such a view.

We present in this talk an alternative approach making it possible to describe separately any information, whatever its level or granularity. It is then possible to describe linguistic phenomena in terms of interaction between different sources of information. The solution consists of representing all kinds of information by means of constraints, the parsing technique being a constraint satisfaction one. It is then possible to characterize precisely the different properties of an utterance. We will present in this perspective a formalism called Property Grammars which implements these aspects.

Short bio:

Philippe Blache is a researcher in computational linguistics. He is the director of the "Laboratoire Parole et Langage" (University of Aix-en-Provence, France). His researches focus on the implementation of linguistic theories and the development of constraint-based solutions for natural language processing. He is a member of different international boards (European chapter of the ACL, ESSLLI standing committee).

Parking: Visitors requiring a parking pass are asked to contact us at least one working day before the seminar.

Enquiries: sals@ics.mq.edu.au