Trait impressions as overgeneralized responses to adaptively significant facial qualities: Evidence from connectionist modeling

Connectionist modeling experiments tested anomalous-face and baby-face overgeneralization hypotheses proposed to explain consensual trait impressions of faces. Activation of a neural network unit trained to respond to anomalous faces predicted impressions of normal adult faces varying in attractiveness as well as several elderly stereotypes. Activation of a neural network unit trained to respond to babies' faces predicted impressions of adults varying in babyfaceness as well as 1 elderly stereotype. Thus, similarities of normal adult faces to anomalous faces or babies' faces contribute to impressions of them quite apart from knowledge of overlapping social stereotypes. The evolutionary importance of appropriate responses to unfit individuals or babies is presumed to produce a strong response preparedness that is overgeneralized to faces resembling the unfit or babies.