Abstract
Vision scientists have long argued that the perception of faces is particularly special among other visual categories, due to its configural and holistic nature. Humans also have extensive experience with faces, likely more than with any other visual category, and this has led to investigations into expertise, which often use stimuli that are learned within the laboratory. However, these novel stimuli possess numerous features such as shape, curvature, or orientation, and these features alone are sufficient for category learning in these experiments. We therefore designed stimuli for category learning experiments which can only be learned from configural information. The stimuli within each category share exactly the same features; the only information that differentiates them is how they are configured. We also created matched featural-stimuli, where configuration is not relevant; the only information that differentiates them is which features are present. Participants learned category membership through feedback learning. First, we varied the dimensionality of the feature space which defines the categories, and found that most participants are able to learn configural categories from a feature space with 3 dimensions within 100 trials. While some subjects were able to learn higher dimensional configural categories, most were unable to learn configural categories beyond 3 dimensions of features, even after 2000 trials. Next, we tested some well-known phenomena observed in face perception, and found that the configural-stimuli, but not the featural-stimuli, follow a similar pattern of results as faces. Configural-stimuli produce a marked and significant inversion effect in reaction time, while featural-stimuli did not. Furthermore, the configural-stimuli produce a strong misalignment effect, while featural-stimuli did not. Overall, we present a novel framework for studying configural and holistic processing in the absence of feature-driven effects or expertise, which would otherwise confound or interact with research into the mechanisms of configural processing.