Context-Aware Conversational Agents Using POMDPs and Agenda-Based Simulation

Abstract

Abstract. Context-aware systems in combination with mobile devices offer new opportunities in the areas of knowledge representation, natural language processing and intelligent information retrieval. Our vision is that natural spoken conversation with these devices can eventually become the preferred mode for managing their services by means of conversational agents. In this paper, we describe the application of POMDPs and agenda-based user simulation to learn optimal dialog policies for the dialog manager in a conversational agent. We have applied this approach to develop a statistical dialog manager for a conversational agent which acts as a voice logbook to collect home monitored data from patients suffering from diabetes.