A Taxonomy of Empathetic Questions in Social Dialogs

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 2952 - 2973 May 22-27, 2022 c⃝2022 Association for Computational Linguistics A Taxonomy of Empathetic Questions in Social Dialogs Ekaterina Svikhnushina, Iuliana Voinea, Anuradha Welivita and Pearl Pu School of Computer and Communication Sciences EPFL, Lausanne, Switzerland {ekaterina.svikhnushina,iuliana.voinea, kalpani.welivita,pearl.pu}@epfl.ch Abstract Effective question-asking is a crucial compo- nent of a successful conversational chatbot. It could help the bots manifest empathy and ren- der the interaction more engaging by demon- strating attention to the speaker’s emotions. However, current dialog generation approaches do not model this subtle emotion regulation technique due to the lack of a taxonomy of questions and their purpose in social chitchat. To address this gap, we have developed an empathetic question taxonomy (EQT), with special attention paid to questions’ ability to capture communicative acts and their emotion- regulation intents. We further design a crowd- sourcing task to annotate a large subset of the EmpatheticDialogues dataset with the estab- lished labels. We use the crowd-annotated data to develop automatic labeling tools and pro- duce labels for the whole dataset. Finally, we employ information visualization techniques to summarize co-occurrences of question acts and intents and their role in regulating inter- locutor’s emotion. These results reveal impor- tant question-asking strategies in social dialogs. The EQT classification scheme can facilitate computational analysis of questions in datasets. More importantly, it can inform future efforts in empathetic question generation using neural or hybrid methods.1 1 Introduction Questions constitute a considerable part of casual conversations and play many important social func- tions (Huang et al., 2017; Enfield et al., 2010). Asking follow-up questions about the speaker’s statement indicates responsiveness, attention, and care for the partner (Bregman, 2020; Huang et al., 2017). Listeners who manifest such an empathetic and curious attitude are more likely to establish the common ground for meaningful communication 1Our code and the annotated dataset are publicly accessible at https://github.com/Sea94/EQT. (McEvoy and Plant, 2014) and appear more likable to the speakers (Huang et al., 2017). The vital role of questions in social interaction makes question-asking a desirable property for open-domain chatbots. These chatbots aim to en- gage in a natural conversation with the users while practicing active listening to deliver understanding and recognition of users’ feelings (Rashkin et al., 2019). In fact, generating meaningful questions is so important that this has become one of the central objectives of such agents (Xiao et al., 2020). However, asking questions effectively is chal- lenging as not all questions can achieve a particular social goal, such as demonstrating attentiveness or empathy (Huang et al., 2017; Robinson and Heritage, 2006; Paukert et al., 2004). Given the task complexity, automatic conversational question generation is still gaining momentum, with only few results reported so far. See et al. (2019) sug- gested a way to control the number of questions produced by the model with conditional training. Wang et al. (2019) proposed a question-generation method to increase their semantic coherence with the answer, employing reinforcement …

Content truncated to 500 words