An Approach to Integrating Emotion in Dialogue Management

摘要

Presented in this paper is a method for the construction of emotion-enabled embodied (conversational) agents. By using a modified POMDP model, this method allows dialogue management not only to include emotion as part of the observation of user's actions, but also to take system's response time into consideration when updating belief states. Consequently, a novel algorithm is created to direct conversation in different contextual control modes, whose dynamic changes further provide hints for emotion animation with facial expressions and voice tunes. Experiment results demonstrate that the integration of emotion in dialogue management makes embodied agents more appealing and yields much better performance in human/computer interaction.