Abstract
Models of configural processing often neglect an explanation of how configural mechanisms and representations develop in visual perception. Employing standard information processing models (see Townsend & Ashby, 1983), we investigate how configural learning via perceptual unitization affects and is affected by the underlying information processing system. Over the course of training in a conjunctive categorization task, participants unitized a novel object by perceptually joining individual features into fewer, larger features, leading to a single, holistic representation. Using the capacity coefficient (Wenger & Townsend, 2000), which provides an index of work-load efficiency, we demonstrate that the configural learning process leads to a qualitative shift from extremely limited- to extremely super-capacity processing due to the effective reduction in work load via unitization. Thus, configural learning qualitatively changes the way configural information is processed; additionally, after training, the processing system is consistent with Townsend and Wenger's (2001) working model of configural processing: an exhaustive parallel or coactive system with facilitatory channel interactions which exhibits super-capacity processing. We have modeled these capacity changes with a neurologically-motivated Hebbian learning rule embedded in a parallel system; limited capacity is produced by negative interactions and super capacity by positive interactions (Townsend & Blaha, In Preparation). Alternatively, the shift in capacity could reflect a structural change in architecture from slow, serial processing to fast, parallel processing. Current experiments investigate processing architecture in order to determine which system best captures the essence of configural learning.