How Do We Answer Complex Questions: Discourse Structure of Long-form Answers

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 298 - 307 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Multi-Party Empathetic Dialogue Generation: A New Task for Dialog Systems Lingyu Zhu1, Zhengkun Zhang1, Jun Wang2 Hongbin Wang3, Haiying Wu3, Zhenglu Yang1∗ 1TKLNDST, CS, Nankai University, 2MS, Ludong University, 3Mashang Consumer Finance Co., Ltd., China {zhulingyu,zhangzk2017,junwang}@mail.nankai.edu.cn, {hongbin.wang03,haiying.wu02}@msxf.com, yangzl@nankai.edu.cn Abstract Empathetic dialogue assembles emotion un- derstanding, feeling projection, and appropri- ate response generation. Existing work for em- pathetic dialogue generation concentrates on the two-party conversation scenario. Multi- party dialogues, however, are pervasive in re- ality. Furthermore, emotion and sensibility are typically confused; a refined empathy analysis is needed for comprehending fragile and nu- anced human feelings. We address these is- sues by proposing a novel task called Multi- Party Empathetic Dialogue Generation in this study. Additionally, a Static-Dynamic model for Multi-Party Empathetic Dialogue Genera- tion, SDMPED, is introduced as a baseline by exploring the static sensibility and dynamic emotion for the multi-party empathetic dia- logue learning, the aspects that help SDMPED achieve the state-of-the-art performance. 1 Introduction Empathetic conversation studies have been coming to the forefront in recent years owing to the in- creasing interest in dialogue systems. Empathetic dialogues not only provide dialogue partners with highly relevant contents but also project their feel- ings and convey a special emotion, that is, empathy. As revealed by previous studies (Fraser et al., 2018; Zhou et al., 2020), empathy can enhance conver- sation quality and transmit appropriate emotional responses to partners. Accordingly, most, if not all, existing work focuses on taking an emotional perspective in dialogue studies (Levinson et al., 2000; Kim et al., 2004; Bertero et al., 2016; Fraser et al., 2018; Rashkin et al., 2019). Although the empathetic conversation has re- ceived extensive attention, its exploration is still limited to the scenario with only two parties. In fact, multi-party chatting scenes are common in seminar discussions, conferences, and group chats. *Corresponding author. ﹏﹏ I have been fighting for so long. And I don’t want to talk this with my family or friends. I’m confused and lonely. [Sad] Oh, I am so sorry. It’s too bad for you. [Worried] Don’t worry! [Calm] Your words make me feel better. Thank you so much. Means so much. [Relxed] Weak Sensibility Moderate Sensibility Strong Sensibility ﹏﹏﹏﹏ You are not alone. Just be relaxed, I went through this and I truly understand how bad it is. [Supportive] Gee! Believe you can achieve it! [Optimistic] High Low Sensibility: Emotion: Negative Positive Happy to hear that. Working together, we can make the future better! [Happy] ﹏﹏﹏﹏﹏ Speaker 1 Speaker 1 Speaker 2 Speaker 3 Speaker 4 Speaker 4 Speaker 3 Figure 1: An empathetic dialogue example of multi- party. When people with different sensibilities respond to the same requests for help, their emotions and empa- thy differ. Different shades of red and blue denote the degree of positive and negative emotions, and different shades of green denote the degree of …

Content truncated to 500 words