Understanding what features influence the believability of expressions in computer-generated (CG) faces is key to predicting how humans will respond to them. In human faces, eye-gaze has been shown to influence the interpretation of expressions. The present study advances understanding by testing whether eye-gaze affects the believability of CG people's expressions - how much they appear to come from genuinely-felt emotion. First, participant (N = 70) ratings of believability and emotion clarity were used to identify a set of angry, fearful, happy and sad facial expressions for use in Study 1. Study 1 (N = 150) then measured believability for these CG expressions paired with one of six increasingly sideways (anger, fear) or downcast (happy, sad) gaze aversions. Happiness and anger were most believable with direct rather than averted gaze, while sadness became increasingly believable as gaze turned downward. Fear showed no effect of gaze. Study 2 (N = 64) replicated the increased believability of sadness with downcast gaze but showed decreased believability with sideways aversion. Overall, our results highlight the theoretical importance of alignment between the signalled meaning of co-occurring facial cues in driving perceptions of believability and provide practical guidance on how gaze can optimise the believability of CG facial expressions.
Haile et al. (Tue,) studied this question.