5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

Pacing Spoken Directions to Suit the Listener

Tatsuya Iwase, Nigel Ward

University of Tokyo, Japan

To make human-computer dialog as `natural' as human-human dialog requires paying attention to the timing of utterances. This is done with reference to responses from the listener, in particular back-channel feedback, questions and mumbles. On the basis of corpus analysis, We have made direction-giving dialog system which adjust the pace of dialog using only prosodic information without using speech recognition; no word recognition was used. We contrived a method to evaluate a dialog system talking to human naturally with prosodic information. To evaluate the naturalness of dialog made by our system, we made three experiment with 10 subjects each. The system accomplished natural dialog, and most of subjects weren't aware that it was a computer. This fact, that reasonably good performance was obtained by paying attention to prosodic information alone, indicates the utility of using prosody in producing appropriate timing in dialog. This confirms a commonly held belief.

Full Paper

Bibliographic reference.  Iwase, Tatsuya / Ward, Nigel (1998): "Pacing spoken directions to suit the listener", In ICSLP-1998, paper 0224.