Owls and Wading Birds: Generalization Gradients in Expertise

Recently, Tanaka et al. (in press) have shown that subjects trained at the subordinate level (henceforth, “experts”) exhibit an advantage over subjects trained at the basic level on the same stimuli in performing discrimination tasks within their domain of expertise, showing that “mere exposure” to a category is not enough to induce discrimination behavior consistent with expertise. In addition, experts generalize their discrimination performance in a graded fashion to novel exemplars from known classes, as well as novel exemplars from novel, but related classes. We applied our two-component neurocomputational model of perceptual expertise to this domain (Sugimoto & Cottrell, 2001; Joyce & Cottrell, 2004). To our surprise, we found that we could not match the data with our original model. Ironically, we needed to add a new component that models “mere exposure” in order to account for the discrimination performance on basic level categories.