Thomas E. Nichols, William F. Eddy, James L. McClelland
We use Functional Magnetic Resonance Imaging (fMRI) to test two competing theories of visual perception. Expressed as connectionist models, the two theories make different predictions about the role context plays in visual perception. McClelland and Rumelhard's interactive activation model (1981) asserts that context effects the actual formation of the percept, while the ``bottom up'' model asserts that context is integrated with visual stimuli at higher cognitive levels. Using pairs of ambiguous stimuli (quickly flashed words) preceded by context (sentences), we image the primary visual cortex looking for context-dependent activity. We use a combination of classical and Bayesian statistical tools to model the signal. Four normal controls were studied, two subjects yielded usable data; in just one subject there was context-dependent activity, but not in the manner hypothesized. This result and the small amount of usable data precludes making any solid conclusions about the two theories of visual perception.