Explorations in multiparty casual social talk and its relevance for social human machine dialogue

Much talk between humans is face-to-face, casual, multiparty, and of indefinite duration. Such casual conversation or social talk facilitates social bonding and mutual co-presence rather than strictly being used to exchange information in order to complete well-defined practical tasks. Artificial partners capable of participating as a speaker or listener in such talk would be useful for companionship, educational, and social contexts. However, to adequately model social talk, such applications require dialogue structure beyond simple question/answer routines. While there is a body of theory on multiparty casual talk, there is a lack of quantitative work in the area. Our work focuses on the anatomy of casual talk, in particular phases of chat, highly interactive dialogue exchanges, and chunks, longer contributions from single participants in the dialogue. We outline the current knowledge on the structure of casual talk and describe our investigations in this domain. Our research finds that distributions of the durations of chat and chunk phases vary with chat being shorter than chunk phases. Chat is also more common at the start of conversations, with chunks becoming more prominent as the conversation progresses. Laughter and overlap are more common in chat phases than chunk phases. We discuss how these insights can inform the design and implementation of truly social machine dialogue partners.