Combining Several User Models to Improve and Adapt the Dialog Management Process in Spoken Dialog Systems

Abstract

Spoken dialog systems have demonstrated a high potential for more flexible, usable and natural human-computer interaction. These improvements are highly dependent on the users' adaptation and dialog management processes, which respectively integrates adaptation capabilities and decides the next system response for the current dialog state. In this paper we propose to carry out the fusion of the user's adaptation and dialog management processes at the dialog level in a single step. To do this, we describe an approach based on a statistical model that combines two models for user's intention modeling, generates a single representation from the users utterances and their confidence scores, and selects the next system action based on this representation. The paper also describes the practical application of the proposed approach to develop a dialog system providing travel and tourist information.