Abstract
How does category learning affect visual perception of objects? While there is agreement that category learning improves the ability to later discriminate between objects, there is some debate whether this improvement is selective for relevant object dimensions or not. In prior work (Folstein et al., submitted), we noted that studies finding selective improvements along relevant dimensions used “dimensional” morphspaces, constructed by morphing between separate morphlines defining an x-axis and a y-axis, while studies showing no selective improvements used “polar” morphspaces, created by morphing directly between morphparents. One interpretation of these results is that dimensional spaces have separable dimensions even before any category learning, corresponding to the morphlines used to construct them, while polar spaces have integral dimensions. Another possibility is that both dimensional and polar spaces are have integral dimensions before any category learning, but that category learning causes dimensional spaces to become separable through a process known as differentiation. We first trained participants to categorize objects in dimensional or polar morphspaces according to diagonal or orthogonal category boundaries. Participants were equally accurate in learning diagonal and orthogonal boundaries in both spaces, suggesting that both spaces are initially integral. We next trained those participants to categorize the same spaces according to a new category boundary that was rotated by either 90 degrees or 45 degrees relative to the original boundary. Now learning in the dimensional space was worse when the boundary had been rotated 45 degrees than when it had been rotated 90 degrees, suggesting that the initial category learning caused differentiation into arbitrary separable dimensions orthogonal to the learned category boundary (Goldstone & Steyvers, 2001). In contrast, there was no evidence of differentiation in the polar space. These results provide further evidence that morphspace structure places important constraints on the effect of category learning on visual perception.
Supported by a grant to the Perceptual Expertise Network from the James S. McDonnell Foundation, the Temporal Dynamics of Learning Center (SBE-0542013), an NSF Science of Learning Center, NEI award 2 R01 EY013441 to IG, a grant from the Vanderbilt Vision Research Center (P30-EY008126), and NIE NRSA award 1 F32 EY019445-01 to JRF.