Abstract
A recognition advantage for distinctive versus typical faces has been widely reported (e.g., Valentine, 1991, Deffenbacher et al., 2000). The effect is robust and appears as both an increased hit rate for distinctive targets and a reduced false alarm rate for distinctive distractors. Because distinctive faces lie in a sparse, peripheral region of face space, the recognition advantage could potentially be due entirely to the fact that they are more dissimilar than typical faces to any randomly chosen set of distractor faces. To control for target-distractor distances, we constructed a parameterized space of silhouetted face profiles. A previous set of studies validated the parameterized silhouettes as genuine face stimuli. In this study, we used a 3-AFC delayed match-to-sample paradigm to test whether distinctive faces retain their processing advantage over typical faces when target-distractor distances are matched for the two types of faces. In the first condition, a set of typical and distinctive silhouettes were constructed to lie on concentric hyper-spheres in silhouette face space. In the second condition, typical and distinctive silhouettes were constructed to occupy regions of equal size in silhouette face space. In both conditions, the recognition advantage normally associated with distinctive items disappeared. In fact, we observed a recognition disadvantage for distinctive silhouettes. We consider an simple explanation for this “reverse distinctiveness effect” in terms of norm-based coding and perceptual learning.
Gordon Bower, Tom Griffiths, Doug Medin, Mark Steyvers, Josh Tenenbaum, Dan Yarlett