Attention warps visual-semantic representations across the human cerebellum

Abstract

Humans effortlessly shift their attention between objects and events in the environment. These attentional shifts modulate the representation of information in many regions of the cerebral cortex, including those that represent visual-semantic information. Based on recent evidence that the cerebellum also represents semantic information, we hypothesized that attention would also shift semantic representations in the cerebellum. To test this hypothesis, we used fMRI to measure brain activity while participants watched naturalistic movie clips and were instructed to attend to either people or vehicles. We used a voxelwise encoding model framework to recover visual-semantic representations in the cerebral cortex and cerebellum under the two conditions, and quantified the direction and magnitude of visual-semantic tuning shifts between conditions. Attention to different semantic categories had a large effect on visual-semantic tuning in the cerebellum. Representations in some regions shifted toward the attended category, and representations in other regions shifted away from the attended category. Within cerebellar regions that best represent visual-semantic information, the tuning shifts were most pronounced in Crus I and II and in lobule VI. Further, the magnitude of attentional tuning shifts was greater in the cerebellum than in the cerebral cortex, particularly for tuning shifts towards the attended category. These differences suggest that attentional modulation in the cerebellum is different than in the cerebral cortex and may not solely be driven by cerebral inputs.