Scaffolding the social mind: emotion recognition supports mentalizing but reinforces bias

Mentalizing-inferring others' mental states-is a core social function supported by both perceptual and contextual processes. Emotion recognition similarly unfolds in two stages: early perceptual decoding of facial features followed by context-sensitive interpretation. Here, we tested whether emotion recognition scaffolds mentalizing across these stages and whether they attenuate or amplify social bias. Participants (N = 34) completed a face categorization task during EEG recording, in which mentalizing was operationalized as differential neural responses to mindful (human) versus mindless (doll-like) faces, while perceptual (neutral vs. smiling) and contextual (ingroup vs. outgroup) cues were systematically manipulated. Early neural responses differentiated human from doll-like faces, indicating rapid mind detection despite perceptual similarity. Smiling expressions enhanced early neural sensitivity across both human and doll-like faces, suggesting that emotional information is extracted at early perceptual stages independently of stimulus humanity. Later EEG components revealed an asymmetry: emotional expressions selectively boosted mental state attribution for ingroup faces, reflecting the integration of contextual social information at later stages. Implicit association tests confirmed this pattern, with mind-related concepts more strongly associated with smiling ingroup faces.