Abstract
Action recognition relies on a network of prefrontal, parietal and middle temporal brain regions. A hierarchy of action representation in these regions is observed from specific perceptual and kinematic features such as movements of body parts or modality-specific sensory representations to more stimulus-general, conceptual aspects. However, the functional profiles of these regions remain unclear. Specifically, it is typically assumed that the "action recognition system" refers to a network of regions that supports recognizing actions of humans (e.g., a person jumping over a box). Yet, inanimate entities can also be involved in motion events that are structurally similar to human-agent actions (e.g., a ball bouncing over a box). Here, we used fMRI-based MVPA to test which components of the action recognition system encode action representations that are specific to goal-directed actions of human agents or more general representations that also define structurally similar non-agentive motion events. During fMRI, participants observed structurally similar human-agent actions and inanimate events. Cross decoding revealed that large parts of prefrontal, parietal and middle temporal cortices carry similar representational profiles in encoding meaningfully different human-agent actions or non-agentive motion events. Furthermore, a subregion in right superior temporal sulcus could better distinguish human-agent actions compared to non-agentive actions. These findings imply that action representations that are encoded in these frontoparietal and middle temporal regions cannot be limited to sensorimotor features specific to human-agent actions. However, the action recognition system also contains components that are distinctly associated with human-agent motion processing.