Stefan Riezler

Graduiertenkolleg ILS, Seminar für Sprachwissenschaft, Universität Tübingen, Wilhelmstr. 133, 72074 Tübingen, Germany,
phone: ++49 7071 2972732
e-mail: riezler@sfs.nphil.uni-tuebingen.de(riezler@sfs.nphil.uni-tuebingen.de) ,

Supervisors
Prof. Dr. Erhard Hinrichs
Dr. Steven Abney

Preliminary Title
Probabilistic Constraint Logic Programming:
Formal Foundations and Applications to Natural Language Grammars

Abstract
Constraint logic grammars provide a powerful formalism for complex logical descriptions of natural language phenomena in exact terms. Describing some of these phenomena may, however, run counter to the restrictive exactness of classical logics and instead require some form of graded distinctions which are not provided by such grammars.
Since the advent of statistical methods in computational linguistics, much work has been done on probabilistic versions of regular and context-free grammars. In addition to many other applications, these approaches provide graded distinctions, e.g., for the general problem of ambiguity resolution. Here probabilistic parameters can be automatically estimated from natural language corpora and allow to distinguish parses by their respective probabilities. Parsing can be speeded up by pruning low probability subparses.
The aim of this thesis is to apply some of these methods to the framework of highly expressive constraint logic grammars such as those underlying HPSG. On the one hand, we want to maintain the formal semantic basis of constraint logic grammars. The key in this context is to provide a formal semantics for weighted constraint logic grammars in a framework quantitative constraint logic programming. On the other hand, we want to develop on this basis a framework for probabilistic constraint logic grammars. The key in this context is to define a probabilistic model for constraint logic programming which allows for automatic parameter estimation and best-parse search.