Not all spaces stretch alike: How the structure of morphspaces constrains the effect of category learning on shape perception

摘要

How does the way we experience and categorize the world affect the way we visually perceive the world? By some perspectives, visual representations provide input for categorization but are not significantly altered by categorization. Others argue that perception is required for categorization but that categorization also alters visual perception. The latter view is supported by studies showing that visual features of categorized objects become more discriminable following category learning, but only if the features are useful or a??diagnostica?? for categorization. Evidence for this phenomenon is mixed, however. We investigate an explanation that has remained unexplored up to now: the structure of the morphspaces categorized by participants. Studies that do not find increases in discriminability often use a??polara?? morphspaces, with morph-parents lying at corners of the space, while studies with positive results use a??dimensionala?? spaces, defined by orthogonal morphlines, each a dimension created by morphing two parents. Using the same four morph-parents, we created dimensional and polar morphspaces matched in mean pair-discriminability. Categorization caused a selective increase in discriminability along the diagnostic dimension of the dimensional space, but not the polar space. This suggests that polar morphspaces should be used if one wishes to avoid selective increases in perceptual discriminability caused by categorization but dimensional morphspaces should be used if one is interested in the effect of selective attention to object properties. In addition, our results suggest that previous fMRI and electrophysiological studies finding little effect of category learning in the visual system (as well as modest behavioral effects on perception) may have been limited by the use of polar spaces.