- Realizing Timing-free Utterances for Spoken Dialogue Systems -
Katsuhiko SHIRAI
School of Science and Engineering, Waseda University
3-4-1 Okubo, Shinjuku-ku, Tokyo, Japan
shirai@shirai.elec.waseda.ac.jp
Aiming at ensuring the freedom of utterance timing, that is system allows users to utter any time they like, the handling methods of user interrupts are discussed. Ensuring of freedom of user is very important to realize natural dialogue between human and machine. The constraints against speaking contents, expression and also speaking timing prevent users from their natural thought and affect bad influence to the communication. As for the efforts to remove the constraints against speaking contents and expression, several study have been tried. However, there is no systematic approach for the timing problem. In this study, we focus on this timing problem. To deal with timing problem, the control of the speaker's right is essential. We collect the dialogue corpus and observed them for finding the cue for the switch of speaker's right. In this paper, we classified the types of the moves of speaker's right and defined the cue for each type. When the system allows users to interrupt, another big problem arise. Conventional system which is not deal with interrupt usually adopt sentences as the unit of planning of utterance. In this case however there must be difference between the contents the system intended to inform and that user received. To solve this problem, we proposed a new utterance planning method. In this method, the items to be informed in a sentence are adopted as the unit of planning. Using this method, the time required to complete a task is reduced and the active utterances based on their own idea. Thus, a user friendly speech conversation system is realized.
Keywords: Spoken Dialogue, Utterance timing, Interruption, Turn taking