Understanding conversational interaction in multiparty conversations: the EVA Corpus

Abstract This paper focuses on gaining new knowledge through observation, qualitative analytics, and cross-modal fusion of rich multi-layered conversational features expressed during multiparty discourse. The outlined research stems from the theory that speech co-speech gestures originate same representation; however, representation is not solely limited to production process. Thus, nature how information conveyed by synchronously fusing must be investigated in detail. Therefore, this introduces an integrated annotation scheme methodology which opens opportunity study verbal (i.e., speech) non-verbal visual cues with a communicative intent) components independently, still interconnected over common timeline. To analyse interaction between linguistic, paralinguistic, discourse help improve natural language generation embodied agents, high-quality multimodal corpus, consisting several layers spanning syntax, POS, dialogue acts, markers, sentiment, emotions, behaviour, gesture units was built represented It first its kind for Slovenian language. Moreover, detailed case studies show tendency metadiscourse coincide behaviour non-propositional origin. analysis further highlights newly created model corresponding information-rich consistent corpus can exploited deepen understanding