Analysis of Group Conversations: Modeling Social Verticality

Social interaction is a fundamental aspect of human life and is also a key research area in psychology and cognitive science. Social psychologists have been researching the dimensions of social interaction for decades and found out that a variety of social communicative cues strongly determine social behavior and interaction outcomes. Many of these cues are consciously produced, in the form of spoken language. However, besides the spoken words, human interaction also involves nonverbal elements, which are extensively and often unconsciously used in human communication. The nonverbal information is conveyed as wordless messages, in parallel to the spoken words, through aural cues (voice quality, speaking style, intonation) and also through visual cues (gestures, body posture, facial expression, and gaze) [31]. These cues can be used to predict human behavior, personality, and social relations. It has been shown that, in many social situations, humans can correctly interpret the nonverbal cues and can predict behavioral outcomes with high accuracy, when exposed to short segments or “thin slices” of expressive behavior [1]. The length of these thin slices can change from a few seconds to several minutes depending on different situations. Computational analysis of social interaction, in particular of face-to-face group conversations is an emerging field of research in several communities such as human–computer interaction, machine learning, speech and language processing, and computer vision [20, 38]. Close connection with other disciplines including psychology and linguistics also exist in order to understand what kind of verbal