Ula McCarthy¹ (), Emily Martin¹, Emmanuella Sanchez¹, Fabian Soto¹; ¹Florida International University

The Other-Race Effect (ORE)—the tendency to individuate own-race faces more accurately than other-race faces—is often attributed to differences in perceptual experience that shape the dimensions observers rely on when encoding facial structure. Understanding how early experience modifies the representation of faces from different races requires stimuli with precise control over three-dimensional facial morphology and analytic tools sensitive to subtle shifts in perceived similarity. This study uses parametrically generated “alien” races, constructed from anthropometric principles to capture structural variation found in natural human groups, to test how different learning histories influence the information observers extract when identifying a face. Participants experienced one of three training histories: identity learning, in which they learned to identify and name 25 faces from Race A; categorization learning, in which they learned to categorize faces as belonging or not belonging to Race B; or no training for Race C. Before the reverse-correlation task, all participants were familiarized with a single target identity to isolate the morphological information supporting identity judgments across conditions. Participants then completed a reverse-correlation identity task in which, across 500 trials per race, they selected which of two randomly perturbed faces more closely resembled the target. Individual classification vectors estimated each parameter’s contribution to perceived identity similarity. Permutation-based cluster tests were applied to determine which parameters significantly contributed to each race-specific template and to evaluate contrasts across learning histories. Across all races and conditions, participants produced robust, highly structured templates that deviated strongly from noise, demonstrating reliable recovery of the information guiding identity judgments. However, contrasts between identity-trained, category-trained, and control templates revealed minimal feature-level differences across conditions, despite clear recognition-memory differences resembling an ORE. These findings indicate that distinct learning histories can modulate recognition performance without producing large shifts in the specific facial features observers rely on to represent identity.

Acknowledgements: This research was funded by NSF grant 2319234 awarded to Fabian Soto.