In the study of perceptual categorization, a key distinction is made between integral and separable dimensions. Integral dimensions are often highly unanalyzable, while separable dimensions are highly analyzable and easy to attend in isolation. Little, Wang, and Nosofsky (2016) showed that when trial-by-trial responses are analyzed, a consistent pattern of sequential effects was found in a modified Garner paradigm using integral-dimension stimuli. The present experiments investigate whether these pronounced sequential effects are also found with separable-dimension stimuli. Two experiments using different separable dimensions were conducted. The results indicate that similar patterns of sequential effects were present for separable dimension stimuli, but, unlike for integral dimensions, the effect of a change in the irrelevant dimension in the filtering task was not found. Further, for separable dimensions, the overall pattern of sequential effects did not vary between the Garner tasks (i.e., control, correlated, and filtering). To explain these results, we fit a sequence-sensitive exemplar model and compared the fits of this model to a novel sequence-sensitive feature model, in which only the relevant feature influences the categorization decision. We find that this feature-based model provides a more compelling account of our separable dimension data, while the full exemplar model provides a better account of the integral dimension data. The findings of the present study provide a morecomplete understanding of perceptual categorization and add to the growing body of literature on the prevalence and critical implications of strong sequential effects in cognitive tasks.
Further tests of sequence-sensitive models in a modified Garner task using separable dimensions
