Construction of Responsive Utterance Corpus for Attentive Listening Response Production

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 7244–7252 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 7244 Construction of Responsive Utterance Corpus for Attentive Listening Response Production Koichiro Ito†1, Masaki Murata†2, Tomohiro Ohno†3, Shigeki Matsubara†1,4 †1Graduate School of Informatics, Nagoya University †2National Institute of Technology, Toyota College †3Graduate School of Advanced Science and Technology, Tokyo Denki University †4Information & Communications, Nagoya University †1,4Furo-cho, Chikusa-ku, Nagoya, 464-8601, Japan †22-1 Eisei-cho, Toyota, 471-8525, Japan †35 Senjuasahi-cho, Adachi-ku, Tokyo,120-8551, Japan ito.koichiro.v1@s.mail.nagoya-u.ac.jp, murata@toyota-ct.ac.jp ohno@mail.dendai.ac.jp, matubara@nagoya-u.jp Abstract In Japan, the number of single-person households, particularly among the elderly, is increasing. Consequently, opportunities for people to narrate are being reduced. To address this issue, conversational agents, e.g., communication robots and smart speakers, are expected to play the role of the listener. To realize these agents, this paper describes the collection of conversational responses by listeners that demonstrate attentive listening attitudes toward narrative speakers, and a method to annotate existing narrative speech with responsive utterances is proposed. To summarize, 148,962 responsive utterances by 11 listeners were collected in a narrative corpus comprising 13,234 utterance units. The collected responsive utterances were analyzed in terms of response frequency, diversity, coverage, and naturalness. These results demonstrated that diverse and natural responsive utterances were collected by the proposed method in an efficient and comprehensive manner. To demonstrate the practical use of the collected responsive utterances, an experiment was conducted, in which response generation timings were detected in narratives. Keywords: spoken language corpus, response timing, listening response, narrative speech 1. Introduction The act of narration is a fundamental human require- ment, and narrating can only be established when there is a listener. However, in Japan, the number of single- person households, particularly among the elderly, is increasing (Statistics Bureau of Japan, 2015; National Institute of Population and Social Security Research, 2018), and there are many situations that listeners can- not be present. Therefore, it is important to increase opportunities for people to narrate. To address this issue, conversational agents, e.g., com- munication robots and smart speakers, are expected to listen to narratives. For these agents to be recognized as listeners of narratives, they have to be functionally able to 1. attentively listen to narratives, and 2. convey that they have attentively listened to the narratives. Function 1 is realized using speech recognition and un- derstanding technologies. An explicit means of real- izing function 2 is providing responses to narratives. In particular, producing gestures and utterances to nar- rative is effective. Here, an utterance to realize func- tion 2, i.e., an utterance made in response to a nar- rative to convey attentive listening, is referred to as an attentive listening response. Representative atten- tive listening responses are backchannels. In terms of backchannels, data collections and analyses have been performed (Ward and Tsukahara, 2000; Kamiya et al., 2010; Yamaguchi et al., 2016), and backchannel gener- ation methods have been proposed (Noguchi and Den, 1998; Cathcart et al., 2003; Fujie et al., 2004; Kitaoka …

Content truncated to 500 words