Integration of conceptual graphs and government-binding theory

Abstract

While much research in natural-language processing (NLP) has been devoted to microlevel analyses of constrained text, many applications, such as machine translation, message understanding and information retrieval, call for capabilities in the understanding of unconstrained text. The paper discusses on step toward this type of NLP system: the integration of Chomsky's government-binding (GB) theory of syntax with Sowa's conceptual-graph (CG) theory of knowledge representation. GB theory provides a parsing technology that surpasses that of phrase-structure grammars, and the CG theory offers a formalism that is suitable for handling natural-language semantics and pragmatics. Their marriage is most natural and synergistic. Not only can their respective strengths be enjoyed, but also most intermediate steps required to build CGs from parse trees can be eliminated, because of the fact that, when it is done independently, a great deal of common knowledge is required both for generating parse trees with a GB-based parser and for translating a parse tree into a CG representation.