DraDDP: A Multimodal Multi-Party Dialogue Discourse Parsing Dataset
Researchers have introduced DraDDP, a new multimodal dataset designed for multi-party dialogue discourse parsing. This dataset, derived from American TV dramas, includes over 6,000 utterances and 9 hours of video, addressing limitations of previous studies that focused on single modalities or two-party conversations. Experiments show that incorporating multimodal information significantly enhances the understanding of dialogue structures and relation types. AI
IMPACT Enables more nuanced understanding of complex, multi-speaker conversations by integrating visual and auditory cues.