Generative Classification of Elements of Nonverbal Communication

Abstract

We propose to use a generative model to classify nonverbal communicative movements (gestures, head and body postures, facial expressions). This model represents the communicative movement as a function of the emotional state, the communicative goal, and reference, and demonstrates the pathways of how the movement could be generated in communication. The proposed generative model distributes communicative movements based on the stimulus for their generation. We distinguish movements (a) related to the meaning of the sentence, (b) emotional uncontrolled movements and their controlled variants, (c) emotional states, simulated to influence the addressee, and (d) movements related to utterance production, expectation of feedback, or lack of feedback. This model can potentially classify the entire spectrum of nonverbal actions and can be applied to control robots and emotional computer agents.