Limits of generalization between categories and implications for theories of category specificity

Abstract

Both domain-specific and expertise accounts of category specialization assume that generalization occurs within a domain but not between domains. Yet it is often difficult to define the boundaries and critical features of object domains. Differences in how categories are defined make it difficult to adjudicate between accounts of category specificity and may lead to contradictory results. For example, evidence for whether car experts recruit the fusiform face area is mixed, and this inconsistency may be due to the inclusion of antique cars in one of those previous studies (e.g., Grill-Spector, Knouf, & Kanwisher, 2004). The present study tested the generalization of expertise from modern to antique cars and found that modern-car experts showed expert discrimination and holistic processing of modern cars but not of antique cars. These findings suggest that the neural specialization underlying perceptual expertise is highly specific and may not generalize to distinct subclasses, even when they share some degree of perceptual and conceptual features.

Figure 1

Mean discrimination sensitivity of modern and antique cars in the expertise task. Error bars are 95% confidence intervals based on the MSE from the omnibus ANOVA. The line shows the cutoff for the expertise criterion.

Figure 2

Example trial structure of the composite task, showing study stimulus and four possible test stimuli. Only aligned top trials are represented here. Test stimuli were also misaligned on half of the trials, and car bottoms were also cued on half of the trials.

Figure 3

Mean congruency effects for sensitivity in the composite task. Error bars represent the 95% CI based on the MSE for the 3-way interaction from the omnibus ANOVA.

Figure 4

Mean congruency effects for response times in the composite task. Error bars represent the 95% CI based on the MSE for the 3-way interaction from the omnibus ANOVA.