Emotion Categories are Represented by a 2-Dimensional Valence-Arousal Space

Abstract

A long-standing debate in the field of facial expressions is whether expressions of emotion are perceived as discrete categories – e.g., happy, surprise, fear, disgust, angry and sad – or according to a lower-dimensional set of continuous variables such as valence and arousal. This unresolved debate continues to hinder knowledge advances on the fundamental nature of facial expression perception. Here, we address this debate by showing that a 2D valence-arousal space of face movements accurately represents discrete emotions. First, we built a 2D valence-arousal space of face movements. We asked forty participants to rate the valence and arousal (on a 5-point Likert scale) of 1,200 facial animations composed of randomly sampled face movements (Action Units – AUs). We ascribed the face movements to the 2D (5 x 5) coordinates of their valence and arousal ratings (see Fig S1). Next, we mapped the six classic emotion categories (using 60 models for each, Jack et al., 2012), by correlating their movements with those of the 5 x 5 coordinates of the 2D valence-arousal space. This showed that emotion categories mapped onto distinct valence-arousal regions. Finally, we validated the positions of the emotion categories in valence-arousal space. We asked 20 new participants (10 females, mean age = 22.6 years) to rate the valence, arousal and emotion category of 1,200 new random facial animations. The analyses revealed that the emotion categories map onto similar regions on the valence x arousal space of face movements. Together, our results show that a 2D valence-arousal space of face movements can accurately represent discrete emotion categories and thereby bridge the gap between opposing dimensional and categorical theories of facial expression perception.