Timing Detection for Realtime Dialog Systems Using Prosodic and Linguistic Information

If a dialog system can respond to the user as reasonable as a human, the interaction will become smoother. Timing of response such as backchannels and turn-taking plays important role in such a smooth dialog as in human-human interaction. We are now developing a dialog system which can generate response timing in real time. In this paper, we introduce a response timing generator for such a dialog system. First, we analyzed conversations between two persons and extracted prosodic and linguistic information which had effects on the timing. Then we constructed a decision tree to detect the timing based on the features coming from the information and examined the decision rules. We also applied the decision tree to a timing generator. The timing generator decides the action of the system at every 100ms in user’s pause. We evaluated the timing generator by subjective and objective evaluation.