Insertion of interjectory response based on prosodic information

This paper investigates `aizuchi´ responses (back-channel feedback, interjectory responses or oral nodding) to user´s utterances by a computer. In telecommunications, `aizuchi´ responses are quite important to inform the hearer´s condition to the speaker. In a man-machine telecommunication environment, user´s comfortableness may improve if a dialog system can give suitable `aizuchi´ to user´s utterances. For this purpose, it is important that a system detect conditions of `aizuchi´ insertion and insert one at appropriate timing. Firstly, occurances of `aizuchi´ were analyzed using a simulated dialog corpus of telephone-shopping-tasks. This showed that there are some prosodic cues in utterances in which `aizuchi´ is inserted. Then, timing of `aizuchi´ insertion was investigated and perceptual experiments were carried out. Lastly, we made a system which gives `aizuchi´ using prosodic information. This system detected suitable positions which agreed with original `aizuchi´s with 77% correct rate