Incorporating Communication Style and Interaction of Speakers for Sarcasm Explanation in Dialogue

Sarcasm Explanation in Dialogue (SED) task aims to uncover the underlying meaning of sarcastic expressions in multimodal dialogues. While previous studies have largely focused on modeling dialogue content, they often neglect the influence of speakers and the interactions between utterances. To address this gap, we propose a novel framework called CISI, which integrates personalized communication styles, inter-speaker interaction relationships, and sarcasm-centric multimodal cues to enhance SED. To capture how personalized styles influence sarcasm expression, we model speakers' communication styles using Satir's Communication Model in psychology. Furthermore, we model the flow of sarcasm through discourse parsing, constructing explicit conversational interaction and dependencies between speakers. Lastly, we design a multimodal fusion module that aligns modality-specific cues with sarcasm-related semantics to enhance understanding. Extensive experiments on the WITS dataset demonstrate that CISI achieves superior performance. We also obtain competitive results on the MUStARD dataset for dialogue-level multimodal sarcasm detection, further showcasing the generalizability of CISI.