Recognition of facial expressions from minimal five-point-light displays

This is the authors' abstract. We don't add key points for this paper.

Facial expression recognition depends on both structural features (e.g., eye shape) and dynamic cues (e.g., muscle movements). However, the relative contribution of these information sources remains unclear, partly due to methodological challenges in isolating kinematic from configural information. Existing point-light display paradigms, while useful, typically employ numerous markers that still enable structural face recovery. Here, we introduce and validate the "Clepsydra" model-a minimal five-point configuration that captures the six basic emotions using markers on the eyebrows (×2), mouth corners (×2), and nose (×1). Stimuli derived from authentic facial expressions were presented to 91 participants in both point-light display (PLD) and full light (FL) conditions. Results showed above-chance recognition for five emotions in the PLD condition: happiness, surprise, anger, disgust, and sadness, while fear remained at chance level. Confusion matrix analyses revealed emotion-specific patterns, reflecting the diagnostic value of specific facial muscle movements (i.e., Action Units, AUs, as defined by the Facial Action Coding System): fear was primarily confused with surprise (sharing upper-face Action Units), while anger and disgust showed reciprocal misclassifications. These findings demonstrate that extremely sparse dynamic information can support emotion recognition, though with marked emotion-specific differences reflecting the diagnostic value of specific Action Units. The Clepsydra model offers a promising tool for isolating kinematic contributions to facial expression processing, with implications for testing sensorimotor simulation accounts and designing future minimal display paradigms.