Principle-Based Processing of Natural Language Using CLP Techniques

Johannes Matiasek

Abstract

This thesis presents a framework for the implementation of principle based grammar formalisms, in particular Head Driven Phrase Structure grammar (HPSG). The achievement of this work is the development of a framework allowing for a declarative and executable specification of principle based grammars employing typed feature structures. Thus it establishes grammar as an operational module independent of processing direction within natural language processing.

The motivation for this work stems from two current trends in computational linguistics: firstly, a shift from phrase structure rule based to principle based grammar formalisms and secondly from the predominant use of typed feature structures to represent linguistic structures.

The availability of well-known techniques for processing phrase structure grammars and the absence of equally well-studied methods in the principle based field have lead to the effect that principle based theories are often recast in a phrase structure based style in order to implement them. In this thesis methods allowing for a direct implementation of principle based formalisms have been developed.

In this work, unification of typed feature structures under the restrictions of principled constraints is interpreted as constraint solving in the CLP paradigm. Principles of grammar, viewed as licensing conditions applying to every node of a feature structure, take the form of conditional constraints. The constraint solving mechanism, implemented in DMCAI-Clp---a variant of SICStus Prolog allowing for unification extensions--- takes care of delaying the enforcement of conditional constraints on insufficiently instantiated feature structures. Thus a declarative specification of the grammar is made possible, which is nevertheless operational independently of the processing direction (i.e. parsing or generation).

The usefulness of this approach to implementing typed feature systems is demonstrated by two applications which are linguistically motivated, but whose integration into existing HPSG implementations is not unproblematic. The direct integration of two-level morphology into an combined sentence-level and word-level HPSG grammar as a relation demonstrates the flexibility of the system. The second application deals with case assignment and argument reduction phenomena in German. It demonstrates how lexical rules, which provoke nonmonotonic effects and thus don't square well with a formalism aiming at declarativeness, can be replaced by a constructive mechanism within the word-level grammar.

Finally compilation techniques and various methods of disjunction handling are discussed as efficiency enhancing extensions to the basic system.

Citation

Matiasek J.: Principle-Based Processing of Natural Language Using CLP Techniques, TU Wien, Dissertation, 1994.

Download

http://www.ofai.at/~johannes.matiasek/papers/diss.ps.gz

Back


Last updated: 2005-06-22, 20:27, © Johannes Matiasek, Wien.