Go to ACL ARR 2024 June homepageA Cognitive-grounded Computational Model for Emotional Alignment During Conversations
Abstract: Aligning emotionally during a conversation means showing appropriate emotional reactions to what our interlocutors say and the emotions they share with us. These appropriate emotional reactions are dictated by social standards and ensure smooth and effective interactions. Based on a psychological framework, and adapting already existing models from cognitive modeling and NLP, we investigate the role in conversational dynamics of 1) social expectations over emotional reactions, 2) internal emotional state, and 3) dialog acts. We implement and compare graph-based models and deep learning models on the task of emotion prediction, employing categorical accuracy as the target metric and using MELD and DialyDialog as benchmarks. The results suggest that the internal emotional state and the dialog acts have an influence on the emotional reaction during conversations. These elements, however, did not show a significant impact within the deep learning model. Possible improvements to the models and insights on future directions are provided.
Paper Type: Long
Research Area: Sentiment Analysis, Stylistic Analysis, and Argument Mining
Research Area Keywords: Computational Social Science and Cultural Analytics, Dialogue and Interactive Systems, Discourse and Pragmatics
Contribution Types: Theory
Languages Studied: English
Submission Number: 2107
Loading