Repairing Conceptual Mismatches in Dialogue: a Computational Approach

In this paper, a computational framework is presented for the generation of elementary speech acts to establish conceptual alignment between a computer system and its user. We clearly distinguish between two phases of the alignment process: message interpretation and message generation. In the interpretation phase, presuppositions are extracted from the user’s message and compared with the system’s semantic model of the application domain, which is represented in Type Theory. Subsequently, in the generation phase, an adequate feedback message is produced in order to resolve detected discrepancies. A conversational strategy is provided that is based upon Gricean maxims and a differentiation of various types of information states of the system, such as private and common beliefs about the domain of discourse, and the goal of the user.