Abstract
The Face Inversion Effect (FIE), characterized by a greater reduction in recognition performance for inverted versus upright faces than objects, suggests that unlike objects, faces are processed holistically – a process disrupted with inversion. Despite many studies investigating FIE in identity recognition, its effect in emotion recognition has shown mixed results. We aim to clarify the effect of FIE on the recognition of different emotions and uncover mechanisms of holistic processing by linking behavioral performance with eye movement. Participants (n=40, White, 30 females, M=21.45 years) completed an expression recognition task of anger, fear, happiness, and sadness for upright and inverted faces with eye-tracking (400 trials per participant). The same face stimuli were presented in a random order with gender, emotion, and orientation counter-balanced across blocks. As expected, participants performed worse in identifying emotions of inverted faces in general. However, while recognition of sadness and anger was worse, recognition of happiness and fear remained unaffected by inversion. Using a data-driven machine-learning-based approach, Eye Movement analysis with Hidden Markov Models (EMHMM), we discovered two representative eye movement patterns adopted by participants during the emotion recognition tasks – eyes-focused and nose-focused patterns. Consistent with literature on diagnostic face regions for identification of different emotions, for upright faces, participants’ eye movement patterns were more eyes-focused for anger, fear, and sadness recognition and more nose-focused for happiness recognition. Interestingly, for inverted faces, participants’ eye movement patterns were only more eyes-focused for fear recognition, while anger, happiness and sadness recognition were more nose-focused. We thus show that the face scanning patterns for different emotions were influenced by the orientation of the face (upright versus inverted), and that FIE and disruption of holistic processing in emotion recognition are modulated by adherence to scanning patterns that obtain the most diagnostic information to identify the expression.