Semantic Inference in Natural Language: Validating a Tractable Approach

Abstract

This paper is concerned with an inferential approach to information extraction reporting in particular on the results of an empirical study that was performed to validate the approach. The study brings together two lines of research (1) the RHO framework for tractable terminological knowledge representation and (2) the Alembic message understanding system. There are correspondingly two principal aspects of interest to this work From the knowledge representation perspective the present study serves to validate experimentally a normal form hypothesis that guarantees tractability of inference in the RHO framework. From the message processing perspective this study substantiates the utility of limited inference to information extraction.