Natural Language for Problem Solving Systems

Abstract : Over the course of the contract, we have had results in three separate projects: natural language interpretation for expert systems (covered under the previous contract), goal oriented explanation generation for expert systems, and generation for intelligent tutoring using domain and explanation knowledge sources (covered under the previous contract and the renewal grant). We have completed implementation of semantics for a natural language interface to expert systems. This has included development of a meta-structure for the underlying rule base derived from linguistic classification of verbs, an algorithm for semantic interpretation using the meta-structure representation, and a facility for completing semantic parses over dialog. We have also run a transportability test, moving the system from the domain of income tax advising to the domain of interview scheduling. In our intelligent tutoring system, we have developed a semantic taxonomy for relations between plans, implemented and tested our plan analyzer, and begun design on flexible interaction between schemas (structures in our explanation knowledge source that control generation of content). Finally, we have completed several evaluation studies of our fully implemented system for generating goal oriented explanations.