Natural Pause Generation in Conversations by Psychologically Plausible Natural Language Processing

2020 Fiscal Year Final Research Report

• Project/Area Number: 18K18504
• Research Category: Grant-in-Aid for Challenging Research (Exploratory)
• Allocation Type: Multi-year Fund
• Review Section: Medium-sized Section 2:Literature, linguistics, and related fields
• Research Institution: Shizuoka University
• Principal Investigator: Kano Yoshinobu 静岡大学, 情報学部, 准教授 (20506729)
• Project Period (FY): 2018-06-29 – 2021-03-31
• Keywords: 自然言語処理 / 間 / 生成

Outline of Final Research Achievements

By adopting a model closer to humans, we aim to generate a natural pause like a humans, and the interrupt/backchannel timing during conversations. We further aim to apply such a spoken language processing to dialogue systems. We recorded a spoken language corpus, then trascribed and assigned annotations of backchannel/interrupt timing, etc. We constructed an inference system using these annotations based on speech and linguistic features. We built a dialogue system and implemented an agent which participates in the natural language division of the annual AI Werewolf Contest, where agents automatically play the conversation game "Mafia".

Free Research Field

自然言語処理

Academic Significance and Societal Importance of the Research Achievements

統計的手法の隆盛により、既存の自然言語処理研究のほとんどは大規模データを取集しやすい書き言葉を前提に研究がおこなわれている。そのために話し言葉の処理性能は未だ不十分なのが現状である。また、人間の言語処理は話し言葉を基盤としている可能性が高く、書き言葉の処理においても話し言葉の処理が有用な可能性があり、「間」をはじめとする音声的な要素も取り込んだシステム構築は、自然言語処理全般に資すると共に、人間の言語処理メカニズムの解明にもつながるものである。