Toward analysis of emotional development using physiological and behavioral data

Emotion is one of the important elements for communication with others. It is expressed by various modalities such as body gestures, changes in the heart rate, and so on. It has been suggested by psychological studies that primitive emotions like pleasure/unpleasure differentiate into six basic emotions. For instance, unpleasure branches into anger, sadness, and fear in the sequence of the developmental process. However, previous studies have focused only on the behavioral data which could be influenced not only by emotion but also the context. This paper proposes an experimental method and a computational model to verify the theory of emotional differentiation. We employ a hierarchical Dirichlet process hidden Markov model which can determine the number of emotional state and deal with multimodal information on emotion. A new contribution of our approach is to analyze not only behavioral information but also physiological information to estimate emotional states more accurately.