Explaining the Variability of Human Nonverbal Behaviors in Face-to-Face Interaction

Modeling human nonverbal behaviors is a key factor in creating a successful virtual human system. This is a very challenging problem because human nonverbal behaviors inherently contain a lot of variability. The variability comes from many possible sources, such as the participant’s interactional goal, conversational roles, personality and emotions and so on, making the analysis of the variability hard. Such analysis is even harder in face-to-face interactions since these factors can interact both within and across the participants (i.e. speaker and listener). In this paper, we introduce our initial efforts in analying the variability of human nonverbal behaviors in face-to-face interactions. Specifically, by exploring the Parasocial Consensus Sampling (PCS) framework [13], we show personality has significant influences on listener backchannel feedback and clearly demonstrate how it affects backchannel feedback. Moreover, we suggest that PCS framework provides a general and effective approach to analyze the variability of human nonverbal behaviors, which would be difficult to perform by using the traditional face-to-face interaction data.