Explaining (Sarcastic) Utterances to Enhance Affect Understanding in Multimodal Dialogues

Conversations emerge as the primary media for exchanging ideas and conceptions. From listener’s perspective, identifying various affective qualities, such sarcasm, humour, emotions, is paramount comprehending true connotation of emitted utterance. However, one major hurdles faced in learning these affect dimensions presence figurative language, viz. irony, metaphor, or sarcasm. We hypothesize that any detection system constituting exhaustive explicit presentation utterance would improve overall comprehension dialogue. To this end, we explore task Sarcasm Explanation Dialogues, which aims to unfold hidden irony behind sarcastic utterances. propose MOSES, a deep neural network takes multimodal (sarcastic) dialogue instance an input generates natural language sentence its explanation. Subsequently, leverage generated explanation understanding tasks conversational setup, sarcasm detection, humour identification, emotion recognition. Our evaluation shows MOSES outperforms state-of-the-art SED by average ∼2% on different metrics, ROUGE, BLEU, METEOR. Further, observe leveraging advances three downstream classification – improvement ~14% F1-score identification recognition task. also perform extensive analyses assess quality results.