Abstract
Recognizing that multiple images belong to the same identity is challenging for unfamiliar faces. We examined the dynamic process of face learning by showing participants multiple images of unfamiliar faces and measuring both likeness ratings and the ability to recognize new images. In Experiment 1, participants (N = 110) learned an identity by rating the “likeness” of 3, 6, or 10 images; recognition of new images of the identity was tested and compared to a no-learning control group. Participants in the rating conditions recognized more new images than controls (p< 0.001), providing evidence of learning. Across four identities and despite the order of images being randomized, likeness ratings decreased linearly in the 3-image condition, p< .001. Higher-order trends were significant in the 6- and 10-image conditions, ps< .01, with dips observed during familiarization. This pattern suggests that building a representation of a new identity involves refining expectations about the range of each person’s variability in appearance (Burton et al., 2016). In Experiment 2 (N = 46) we examined whether expectations extend to the context in which faces are presented. Participants rated the likeness of four repeated images of two identities (presented on consistent background scenes) and four novel images presented on the same background scene vs. a novel, unexpected background. We replicated the likeness-rating effects from Experiment 1 for the first four images; a quadratic trend showed that ratings decreased at the 3rdimage, p< 0.001. Critically, likeness ratings for the four novel images were lower than likeness ratings for the four repeated images only when presented on an unexpected background, p = 0.03. We interpret our findings in light of the predictive coding model (Trapp et al., 2018), which predicts that learning occurs when predictions (e.g., how someone looks, the context in which someone is seen) are violated.