Learning at Different Levels of Abstraction

Previous category learning research and the SUSTAIN (Supervised and Unsupervised STratified Adaptive Incremental Network) model of category learning suggest that preferred category level (in a hierarchy of categories) shifts towards lowerlevel (i.e., more specific) categories when stimuli are perceived to be more distinctive. This shift is in accord with work in expertise. In their domain of expertise, experts excel (relative to novices) at classifying stimuli at lower category levels, but their advantage is attenuated with higher-level categories. The work described here directly tests (within a single study) this predicted interaction between category level and stimulus distinctiveness using well controlled artificial stimuli. The results are consistent with prior work utilizing natural stimuli. The results are also informative for evaluating whether attention is dimension-wide (i.e., all items are represented in a common multi-dimensional space of the same extent) or cluster specific (i.e., different conceptual clusters can stress different stimulus dimensions so that different aspects of different stimuli are stressed). The results suggest that attention is not dimensionwide. Instead, attention can stress different aspects of different stimuli. The implications of these findings for models of category learning are discussed.