A Taxonomy of Empathetic Questions in Social Dialogs

Effective question-asking is a crucial component of a successful conversational chatbot. It could help the bots manifest empathy and render the interaction more engaging by demonstrating 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 crowdsourcing task to annotate a large subset of the Empathetic Dialogues dataset with the established labels. We use the crowd-annotated data to develop automatic labeling tools and produce 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 interlocutor’s emotion. These results reveal important question-asking strategies in social dialogs. The EQT classification scheme can facilitate computational analysis of questions in datasets.

Asking follow-up questions about the speaker’s statement indicates responsiveness, attention, and care for the partner. Listeners who manifest such an empathetic and curious attitude are more likely to establish the common ground for meaningful communication

asking questions effectively is challenging as not all questions can achieve a particular social goal, such as demonstrating attentiveness or empathy

Question acts capture semantic-driven communicative actions of questions, while question intents describe the emotional effect the question should have on the dialog partner. For example, a listener may request information (question act) about the age of speaker’s daughter by asking “How old is she?” after learning about her success with the aim to amplify speaker’s pride of his child (question intent).

We opted for the Empathetic Dialogues (ED) dataset (Rashkin et al., 2019), a benchmark dataset for empathetic dialog generation containing 24,850 conversations grounded in emotional contexts. Each dialog is initiated by a speaker describing a feeling or experience and continued by a listener who was instructed to respond empathetically. The dialogs are evenly distributed over the 32 emotional contexts, covering various speaker sentiments (e.g., sad, joyful, proud). We found the ED dataset to be a rich source of question-asking as over 60% of all dialogs contain a question in one of the listeners’ turns, resulting in a total of 20K listener questions.

Negative rhetoric (1.3%): Ask a question to express a critical opinion or validate a speaker’s negative point without expecting an answer

Positive rhetoric (1.0%): Ask a question to make an encouraging statement or demonstrate agreement with the speaker about a positive point without expecting an answer

Question intents Express interest (57.1%): Express the willingness to learn or hear more about the subject brought up by the speaker; demonstrate curiosity

Express concern (20.3%): Express anxiety or worry about the subject brought up by the speaker

Offer relief (4.8%): Reassure the speaker who is anxious or distressed

Sympathize (3.9%): Express feelings of pity and sorrow for the speaker’s misfortune

Support (2.6%): Offer approval, comfort, or encouragement to the speaker, demonstrate an interest in and concern for the speaker’s success

Amplify pride (2.6%): Reinforce the speaker’s feeling of pride

Amplify excitement (1.9%): Reinforce the speaker’s feeling of excitement

Amplify joy (1.6%): Reinforce the speaker’s glad feeling such as pleasure, enjoyment, or happiness

De-escalate (1.6%): Calm down the speaker who is agitated, angry, or temporarily out of control

Pass judgement (1.6%): Express a (critical) opinion about the subject brought up by the speaker

Motivate (1.0%): Encourage the speaker to move onward

Moralize speaker (1.0%): Judge the speaker.

For example, the question “What happened!?” can be classified as Express interest or Express concern, depending on the valence of the speaker’s emotion.