A Model For Selecting An Initial Discourse Goal Using Plan Recognition

This paper describes a computational model for interactive natural language generation that allows a system to analyze a user's initial query to determine the kind of explanation that is being sought. The system can then select an appropriate goal for the interaction called the discourse goal. This model has been implemented in the Interactive Discourse Planner (IDP) [1, 2]. As a testbed, IDP can plan to give interactive driving route advice or alternatively, driving directions based on semantic information in the user's initial question. This model is currently being extended as part of the design and implementation of B2, an interactive interface to a decision support system that can alternatively quiz, tutor, or advise the user about appropriate medical diagnostic techniques [3]. Keywords: Natural Language, User Interfaces, Computational Linguistics INTRODUCTION Information systems are needed that can engage in different kinds of explanations in natural language. Some explanation..