Who Is Speaking to Whom? Learning to Identify Utterance Addressee in Multi-Party Conversations

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 1909–1919, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 1909 Who Is Speaking to Whom? Learning to Identify Utterance Addressee in Multi-Party Conversations Ran Le13∗, Wenpeng Hu2∗, Mingyue Shang1∗, Zhenjun You2∗, Lidong Bing 4, Dongyan Zhao13 and Rui Yan13† 1 Center for Data Science, AAIS, Peking University, Beijing, China 2 School of Mathematical Sciences, Peking University, Beijing, China 3 Wangxuan Institute of Computer Technology, Peking University, Beijing, China, 4 R&D Center Singapore, Machine Intelligence Technology, Alibaba DAMO Academy {leran.pku}@gmail.com {l.bing}@alibaba-inc.com {wenpeng.hu,shangmy,youzhenjunpku,zhaody,ruiyan}@pku.edu.cn Abstract Previous research on dialogue systems gen- erally focuses on the conversation between two participants, yet multi-party conversa- tions which involve more than two partici- pants within one session bring up a more com- plicated but realistic scenario. In real multi- party conversations, we can observe who is speaking, but the addressee information is not always explicit. In this paper, we aim to tackle the challenge of identifying all the miss- ing addressees in a conversation session. To this end, we introduce a novel who-to-whom (W2W) model which models users and ut- terances in the session jointly in an interac- tive way. We conduct experiments on the benchmark Ubuntu Multi-Party Conversation Corpus and the experimental results demon- strate that our model outperforms baselines with consistent improvements. 1 Introduction As an essential aspect of artificial intelligence, dia- logue systems have attracted extensive attention in recent studies (Vinyals and Le, 2015; Serban et al., 2016). Researchers have paid great efforts to un- derstand conversations between two participants, either single-turn (Li et al., 2016a; Shang et al., 2015; Vinyals and Le, 2015) or multi-turn (Zhou et al., 2016; Yan et al., 2016; Tao et al., 2019a,b), and achieved encouraging results. A more gen- eral and challenging scenario is that a conversa- tion may involve more than two interlocutors con- versing among each other (Uthus and Aha, 2013; Hu et al., 2019), which is known as multi-party conversation. Ubuntu Internet Relay Chat chan- nel (IRC) is a multi-party conversation scenario as shown in Table 1. Generally, each utterance is associated with a speaker and one or more ad- dressees in the conversation. Such a characteristic ∗Equal contribution. † Corresponding author. Table 1: An example of the multi-party conversation in the IRC dataset. Not all the addressees are specified. Speaker Utterance Addressee User 1 ”Good point, tmux is the thing I miss.” – User 1 ”Cool thanks for ur help.” @User 4 User 4 User 2 ”Ahha, you r using something like cpanel.” – User 3 ”Yeah 1.4.0 exactly.” @User 2 User 2 User 4 ”my pleasure :)” – leads to complex speaker-addressee interactions. As a result, the speaker and addressee roles as- sociated with utterances are constantly changing among multiple users across different turns. Such speaker and addressee information could be essen- tial in many multi-party conversation scenarios in- cluding group meeting, debating and forum dis- cussion. …

Content truncated to 500 words