Abstract
Repetition is pervasive in our visual experience. For example, we are more likely to re-encounter an object seen moments ago than a new object. The brain takes advantage of these repetitions to facilitate perception, as indexed by reduced fMRI responses in category-specific ventral visual regions. Such ‘repetition attenuation’ may reflect more efficient neural processing of repeated stimuli. Yet, the mechanisms underlying this repetition effect are controversial. Repetition attenuation may result from stimulus-specific learning revealed when a similar stimulus reappears. Alternatively, repetition attenuation may occur when the reappearance of a stimulus matches expectations for repetition. Here we distinguish these two possibilities by building expectations that two stimuli will be less similar. We performed a rapid event-related fMRI study in which each trial contained three views of a rendered 3D object. In the majority of trials (expected), an object rotated in place through the three views; there was a strong expectation for the third view because it continued the rotation, and because this kind of trial occurred most frequently. In the remaining trials, the third object view was either identical to the second view (similar), or an entirely new object (novel). If repetition attenuation reflects learning, then similar trials should elicit more attenuation. If perceptual expectation is more critical, then expected trials should elicit more attenuation. An object-selective region in the left fusiform gyrus showed robust repetition attenuation for both similar and expected trials, but stronger attenuation for similar trials. Lateral occipital cortex showed reliable attenuation only for similar trials. Left prefrontal regions showed equivalently robust attenuation for both conditions, while other frontal regions tracked expectation. These results suggest that the influence of bottom-up similarity vs. top-down expectation on repetition attenuation differs across brain regions. However, in ventral visual cortex, repetition attenuation is dominated by stimulus-specific learning.
Supported by NIH EY014193.