The effect of variability of context and appearance on face learning

Exposure to multiple unique images of the same face enhances recognition. Using ambient pictures and videos, face-learning studies have shown improved generalization, suggesting more refined memory representations for faces exhibiting greater variability. Although ambient images provide more ecologically valid learning conditions, the relative contribution of different sources of variability remain largely unexplored. In two pre-registered experiments, we investigated the effect of variability in appearance (stable, variable hairstyle) and context (unique, multiple background scenes) during face learning, as these factors may influence the coarse-to-fine encoding of diagnostic facial features. In Experiment 1, eight identities were learned in a blocked manner, whereas they were randomised in Experiment 2. Recognition was assessed using images displaying only inner facial features. We hypothesized that conditions promoting finer encoding of inner facial features (variable appearance and multiple contexts in Experiment 1) or providing stable learning anchors (stable appearance and unique context in Experiment 2) would improve recognition. In Experiment 1, variable appearances enhanced recognition performance relative to stable appearances, despite diverting gaze from inner features during learning. However, Experiment 2, along with supplementary cross-experiment analyses, revealed that randomizing identities eliminated the recognition disadvantage observed for stable faces, possibly by encouraging participants to re-engage attention with each learning image. Contextual variability did not affect recognition performance in either experiment, suggesting that appearance variability and learning procedure are stronger determinants of efficient learning. These findings highlight the critical role of appearance variability and of learning procedure in shaping recognition outcomes.