Content and context in conversations : the role of social and situational signals in conversation structure

Abstract

The increasing growth of multimodal material creates, nowadays, a renewed interest in innovative approaches to information extraction from meetings and multiparty conversations; those approaches make use of various multimodal sources such as video, audio and resulting transcripts from which information can be derived to produce a richer semantic analysis of the social interaction. Social interactions are co-constructed by a combination of contextual and linguistic aspects. However, while the social context of interaction has previously been considered in relation to the emotional level of communication, fewer studies have been devoted to exploring the relationship between the context and the linguistic level of interactions. This thesis proposes a novel view of context in communication, exploring its effect on the discourse structure, independently of the affective and emotional spheres of communication. I explore the influence of two concepts of context on human-human communication: the social context (ensemble of the social signals exchanged by the participants) and the situational context (situation in which the conversation takes place), and show the extent to which both, social and situational contexts, have a discourse function. In both cases, results show that the information carried by their timing can be exploited in the detection of discourse events.