Semantic and Discourse Processing with a Feature Structure-based Production System

This paper presents a production (rulebased) system designed to implement a general purpose discourse processing engine for spoken dialogue systems, and tries to show that the engine can carry out discourse and semantic analyses of various phenomena in spoken language in an integrated manner. The function of the engine is to give semantic/discourse analysis to parsed utterances and to determine proper action for the dialogue system where it is embedded. The production system is equipped with mechanisms for 1) hypothesis handling, 2) simple deduction by class definitions, and 3) lexicon look-up. The core of the semantic/discourse analysis is its discourse object management by which reference resolution is done. The paper will show that the discourse object management mechanism used in the engine can deal with a wide range of spoken language phenomena such as anaphora, cataphora or inversion, ellipsis, and fragmentary input.