2023 Fiscal Year Final Research Report
Analysis of prosodic and syntactic features related to conversation timing using spontaneous conversation corpora
| Project/Area Number |
18K11514
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 61060:Kansei informatics-related
|
|---|
| Research Institution | Institute of Technologists (2022-2023)
National Institute for Japanese Language and Linguistics (2018-2021) |
|---|
Principal Investigator |
Ishimoto Yuichi ものつくり大学, 技能工芸学部, 准教授 (50409786)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
榎本 美香 東京工科大学, メディア学部, 准教授 (10454141)
寺岡 丈博 拓殖大学, 工学部, 准教授 (30617329)
|
|---|
| Project Period (FY) |
2018-04-01 – 2024-03-31
|
| Keywords | 発話末予測 / 統語情報 / 韻律 / 次話者選択 |
|---|
| Outline of Final Research Achievements |
We investigated syntactic and prosodic information related to end-of-utterance prediction using the spontaneous conversation corpus. We defined a model to estimate bunsetsu-phrase order from syntactic and prosodic features. The model parameters were iteratively estimated using the Markov chain Monte Carlo (MCMC) method. The results showed that the number of modifiers whose modifying bunsetsu-phrases have not yet appeared in the middle of a sentence was significant as a syntactic feature regardless of the speaker. However, significant prosodic features varied among speakers. Additionally, it was found that the prosodic features of utterances differed depending on the method of next-speaker selection. These findings suggest that conversational listeners adapt to using predictive features that vary according to each speaker rather than relying on a fixed set of prosodic features.
|
| Free Research Field |
音声コミュニケーション
|
| Academic Significance and Societal Importance of the Research Achievements |
コンピュータとの音声対話により日常生活支援を行う音声アシスタントが用いられるようになってきているが、まだ自然な話者交替をともなう会話は実現できていない。話者交替時に間断なく応答するためには聞き手による話し手の発話末の予測が必要であり、本研究では自発発話音声の統語特徴や韻律特徴を基に発話末予測が可能であるかどうかを調べた。統語情報として係り先未定文節数が発話末までの距離を測る指標となりうることと、韻律情報は発話の性質によって有効性が異なることを示しており、これらによりコンピュータによる自然な話者交替へ実現へと近づいたと考える。
|
