[4Xin2] Poster session 2

[4Xin2-46] Predicting Implicit Emotions Elicited in Conversations

〇Yurie Koga¹, Shunsuke Kando¹, Yusuke Miyao¹ (1.The University of Tokyo)

Keywords:Conversation, Emotion Inference, Emotion Recognition

This paper aims at predicting the elicited emotion by an utterance, which is not always expressed directly in utterances in conversations. Previous studies attempt this task by predicting the emotion annotated to the listener’s next utterance. That is, they assume that the elicited emotion is always expressed in the next utterance. However, people sometimes hide their emotions. Therefore, we propose a new task to directly predict the elicited emotion from an utterance, and develop a base model for it. In addition, we compare the difficulties between this task and other related tasks. Our results showed that our model was reasonably successful in predicting the elicited emotions. Further analysis suggests that dialogue history is important in predicting elicited emotions. It also suggests that, compared to speaker emotions, elicited emotions can be moderately predicted without the listener’s own utterance, and that predicting implicit emotions is more difficult than predicting explicit emotions.