Abstract
The hallmark of categorical perception (CP) is that observers show superior discrimination between equally spaced stimuli on physical continua when those stimuli straddle a category boundary. Recent work has shown that CP effects can be learned. Yet there is debate about whether these learned CP effects are really perceptual and if so about the mechanisms and locus of learning. To address these issues we created morphed continua of cervical cells varying from normal to abnormal. We selected these stimuli because of their novelty to most observers and because they have similar image statistics to natural scenes. In experiment 1 we compared experts and novices. Both groups completed two tasks (order counterbalanced). In one task they categorised the stimuli into normal and abnormal. In the other task they reported whether pairs of stimuli selected from the continua were the same (physically identical) or not. Analysis, using signal detection theory measures, showed that the experts were significantly better at discriminating pairs of stimuli from either side of their category boundary compared to within category pairs. The novices showed no such CP effect. These findings are consistent with the possibility that category learning has modified the experts' perception of the stimuli. In experiment 2 we trained novices to categorise the stimuli, briefly presented to one retinal location, into normal and abnormal. Following training a CP effect was apparent at the trained location, but not at locations 3.6 deg. away. This experiment supports the view that category learning modifies perception of these images. Further the retinotopic specificity of learning suggests that the CP is dependent on perceptual processing at relatively early stages of visual analysis, where smaller receptive field sizes are observed. Thus, we conclude that CP effects can reflect a visual perceptual learning process that is driven by categorisation task and modulated by attention.
Funding for this work was provided by the ESRC