Discussion
Experiment 2 shows two important results. First, the training results suggest that task difficulty may be the main factor in indirect category learning; not category structure.On the one hand, the 1D-Width condition was the easiest and could be learned equally well with direct or indirect categorization tasks. On the other hand, the conjunction rule and the information-integration category structures were more difficult and could only be learned directly. As in Experiment 1, representations of difficult category structures could not be learned indirectly; only representations
of easy category structures could be learned indirectly.However, representations could be learned directly regardless of the category structure. Second, the transfer results suggest that rule-based strategies that were learned directly using classification
could be transferred to the same–different categorization task, but not information-integration strategies that were learned directly using the same classification task. This result is counter-intuitive because it suggests that participants in the information-integration condition know the categories of the individual stimuli, but that they cannot make a sameness judgment.4 Yet, this result is consistent
with previous studies of rule-based and information-integration category structures (e.g., Ashby et al., 2003; Waldron & Ashby, 2001), which suggest that the former are
more abstract (and general), whereas the latter are more procedural (and specific). It was also predicted by the COVIS theory of categorization (Ashby et al., 1998),
which suggests that information-integration category structures are learned using a procedural-based system that relies heavily on a consistent stimulus–response mapping.
The consistent stimulus–response mapping present in the direct categorization task is broken when the participants are transferred to the same–different categorization task