The discourse value of social signals at topic change moments

The dynamics of social interactions during human-human conversation have attracted the attention of many researchers aiming at improving the naturalness of automatic dialogue systems and creating socially aware machines. While many studies have investigate social dynamics in relation to the emotional sphere of a conversation, in this work we investigate whether these dynamics are correlated with the discourse structure, particularly topic changes. We conduct an observational study in two corpora of differing nature. In both corpora we find that at topic terminations, the interactional entropy, defined as the amount of social signals exchanged by the participants, decreases with the introduction of a new topic. We conclude that a relation exists between topic changes and amount of social interaction and, hence, that social signals have a discourse value in addition to a social function. This value, we suggest, is group acknowledgement that a new topic has gained the floor. Understanding these social dynamics within conversations can be exploited in the development of socially aware dialogue systems.