| Project/Area Number |
11610555
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
言語学・音声学
|
|---|
| Research Institution | Japan Advanced Institute of Science and Technology |
|---|
Principal Investigator |
ISHIZAKI Masato Japan Advanced Institute of Science and Technology, School of Knowledge Science, Associate Professor, 知識科学研究科, 助教授 (30303340)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
石崎 雅人 北陸先端科学技術大学院大学, 知識科学研究科, 助教授 (30303340)
|
|---|
| Project Period (FY) |
1999 – 2001
|
| Project Status |
Completed (Fiscal Year 2001)
|
| Budget Amount *help |
¥2,500,000 (Direct Cost: ¥2,500,000)
Fiscal Year 2001: ¥800,000 (Direct Cost: ¥800,000)
Fiscal Year 2000: ¥600,000 (Direct Cost: ¥600,000)
Fiscal Year 1999: ¥1,100,000 (Direct Cost: ¥1,100,000)
|
|---|
| Keywords | Initiative / Context / Annotators' Criteria / Lexical information / Clustering / Maximum entropy method / 談話セグメント / 談話セグメント付与基準 / セグメント単位 / 局所的手がかり / 大局的手がかり / 音声対話コーパス / 談話行為 / 統計的手法に基づく推定 / 索引語頻度・逆文書頻度 / ベイズ決定則 / Nグラム / 対話セグメント / 対話セグメンテーション / 談話分析 |
|---|
| Research Abstract |
Conversation Analysis and Psycholinguistics have revealed that communication is maintained through reciprocal coordination at various levels. For example, under noisy environments like train stations, people speak, up to be heard and ask the others to repeat when they cannot hear. This level is hierarchically organized from attentional (people attend to the others' talk), physical (people can hear what the others say), linguistic (people can understand the words, syntactic structures and semantic contents of the others' talk) and understanding (people can understand the others' utterances in context). To explore reciprocal coordination at the understanding level, this study focused on the concept of dialogue initiative and tackled the problems of 1) how the.boundaries of the initiative can be reliably judged by human annotators and 2) how those boundaries can be predicted by computers. For 1), we created working criteria for judging initiative boundaries and showed that the agreement rate among annotators is around 90 % (κ score is around 07) For 2), we confirmed that the maximum entropy method correctly predicts initiative boundaries using lexical information and block distances calculated based on word clustering at around 70 % recall and 55 % precision.
|
