UC San Diego

Categorical and coordinate stimuli were proposed by Kosslyn (1987) as a set of lateralized visual tasks with a left hemisphere advantage for categorical and right hemisphere advantage for coordinate. A categorical task uses relative positioning to make a judgment, such as the statement that a glass of water is on a table; a coordinate task uses absolute positioning to make a judgment, such as the statement that the glass of water is 3 inches from my hand. Kosslyn hypothesized that categorical tasks depended on low spatial frequencies and coordinate were preferentially processed in higher spatial frequencies (e.g. Baker et al. 1999); however, the literature in subsequent years was inconclusive on this hypothesis (Jager and Postma 2003). Slotnick et al. (2001) directly tested Kosslyn's hypothesis and also arrived at conflicting results. By stratifying by difficulty, they showed that Kosslyn's hypothesis holds only when tasks are difficult enough. Our Differential Encoding (DE) model is a three layer neural network that accounts for lateralization of visual processing via the biologically and developmentally plausible mechanism of differences in the connection spread of long-range lateral neural connections. We first establish certain frequency-encoding properties of the DE model. We then show that our model accounts for Slotnick's psychological data and show Slotnick's analysis does not convincingly explain the conflicting results. Instead, we propose that Kosslyn's initial hypothesis was incorrect: categorical and coordinate stimuli are not differentiated solely by spatial frequencies, which is why lateralization has been inconclusive in the past. These results therefore cannot be captured by models such as Ivry and Robertson's (1998) "Double Filtering by Frequency" model, which is driven directly by lateralization in spatial frequency processing.