著者

秋葉巌 松山洋一 小林哲則

出版者

一般社団法人情報処理学会

雑誌

研究報告音声言語情報処理(SLP)

巻号頁・発行日

vol.2013, no.10, pp.1-8, 2013-07-18

---

多人数会話において発生する発話機会の不均衡を調整するために,機会が十分に与えられず 「置いてけぼり」 状態になっている会話参加者に適切な手続きを経て発話機会を与えることのできる会話ロボットを提案する.特に,ファシリテーションのモデルを扱うための最小単位である 4 者会話 (ロボットを含む) を対象として,主導的に会話を進めている参加者らの状況も考慮しながら,段階的に主導権を奪取し,しかるべき対象者に発話機会を与えるような手続きの計算モデルを提案する.モデル化には,誤りを含むセンサ情報にロバストな部分観測マルコフ決定過程 (POMDP) を用いる.さらに手続きとそのタイミングの適切性について評価した被験者実験の結果を報告する.We propose a facilitation robot harmonizing four-participant conversations. Four-participant conversation is the minimum unit that needs facilitation skills. In general, three is the minimum number of participants of a multiparty conversation. In such three-participant situations, back-and-forth interactions between two participants out of three primarily occur and another participant tends to be left behind, who cannot properly get floors to speak. Here, they need one more participant who helps the participant left behind to harmonize him/her with the others. Conversational robots have potentials to participate in such conversations as the fourth participant. When the robot steps in the situation to help, there should be proper facilitating procedures to obtain initiatives to control conversational contexts. In this paper, we propose a conversational robot system harmonizing four-participant conversations along procedures of obtaining initiatives of topic and floor control. These situations and procedures were modeled and optimized as the partially observable Markov decision process. We conducted an experiment to evaluate appropriateness of the proposal procedures and the result shows evidence of its acceptability.