Structured Inheritance Networks and Natural Language Understanding

Abstract

One of the most salient difficulties facing the designer of a Natural Language Interface is the appropriate characterization of the semantics and pragmatics of new domains. Whereas various syntactic parsers have been developed which are largely domain-independent (at least in principle), the possibility of accomplishing something similar in the realm of semantics and pragmatics has never been convincingly demonstrated. Focusing on the various epistemological bases for associating attributes with entities, we examine some of the requirements that this suggests need be met by a system which claims to provide a domain-independent conceptual schema core, from which domain-specific extensions can be interactively elaborated. In addition to showing how our approach differs from one which tries to arrive at canonical representations in terms of invariant semantic primitives, we sketch out ways in which it can be integrated into a general parsing mechanism.