| Project/Area Number |
08837011
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
談話(ディスコース)
|
|---|
| Research Institution | Kyoto Institute of Technology |
|---|
Principal Investigator |
NIIMI Yasuhisa Kyoto Institute of Technology, Faculty of Eng.and Design, Professor, 工芸学部, 教授 (00026030)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
NISHIMOTO Takuya Kyoto Institute of Technology, Faculty of Eng.and Design, Assistant, 工芸学部, 助手 (80283696)
KOBAYASHI Yutaka Kyoto Institute of Technology, Faculty of Eng.and Design, Associate Professor (o, 工芸学部(平成9年度のみ), 助教授 (40027917)
|
|---|
| Project Period (FY) |
1996 – 1998
|
| Project Status |
Completed (Fiscal Year 1998)
|
| Budget Amount *help |
¥2,200,000 (Direct Cost: ¥2,200,000)
Fiscal Year 1998: ¥700,000 (Direct Cost: ¥700,000)
Fiscal Year 1997: ¥700,000 (Direct Cost: ¥700,000)
Fiscal Year 1996: ¥800,000 (Direct Cost: ¥800,000)
|
|---|
| Keywords | spoken dialog / dialog control strategy / speech recognition errors / confirmation / discourse analysis / internet / reliability of speech recognition / information transmission efficiency / 情報項目 / 認識発話 / 対話インタフェース / 情報の組織化 / 操作発話 / 印象発話 / インタフェースの原則 |
|---|
| Research Abstract |
In this project we emphasized the following three points.
1. Modeling and analysis of dialog control strategies. Current speech recognizers make some errors. So spoken dialog system using such recognizers must confirm what they have recognized or ask its user to repeat utterances when recognition results are not so reliable. Under the assumption that the reliability of recognition results be available, we modeled such dialog control strategies with probabilistic state transition networks and evaluated them by two quantities. The first is the average number of turns taken between a dialog system and its user until an utterance including several information items has been accepted. The second is the probability that all of accepted items have been recognized correctly. We derived the quantitative relation of these quantities to the performance of the speech recognizer used in a spoken dialog system.
2. Discourse analysis in a spoken dialog system. A new method was proposed for discourse analysis in a spoken dialog system. The discourse analysis is important to correctly understanding user' s utterances and to predict what a user is likely to say next. The method was implemented in the spoken dialog system of which the task is a sightseeing guide and proved to successfully work. Furthermore, the proposed method was also used for predicting the content of user' s next utterance and evaluated by using a dialog corpus collected through OZ simulation.
3. A spoken dialog interface to the internet. To develop a spoken dialog interface to access databases on the internet, we collected a corpus of dialogs through OZ simulation and transcribed the dialogs, Based on the relation between words used in the dialogs and internet page images displayed, we developed a dialog control strategy to efficiently retrieve necessary information from the inter-net, and constructed a spoken dialog interface to the internet.
|
