An algorithm for choosing Japanese acknowledgments using prosodic cues and context

In human dialog a wide variety of acknowledgments are used. One function of this seems to be indicating attention, interest, and involvement to the other speaker. Based on study of Japanese memory game dialogs, we propose an algorithm for choosing among acknowledgment responses, including hai (yes), so (right), and un (mm). The primary factors involved are aspects of the user's internal state, including con dence and liveliness, as inferred from the context and the user's prosody. We found that judges preferred algorithm-chosen responses to randomly varied responses, con rming our hypothesis that `sensitive' and subtle choice of response may improve helpfulness and naturalness of man-machine spoken language interaction.