Conceptual knowledge for understanding other's actions is organized primarily around action goals.

Semantic knowledge about objects entails both knowing how to grasp an object (grip-related knowledge) and what to do with an object (goal-related knowledge). Considerable evidence suggests a hierarchical organization in which specific hand-grips in action execution are most often selected to accomplish a remote action goal. The present study aimed to investigate whether a comparable hierarchical organization of semantic knowledge applies to the recognition of other's object-directed actions as well. Correctness of either the Grip (hand grip applied to the object) or the Goal (end-location at which an object was directed) were manipulated independently in two experiments. In Experiment 1, subjects were required to attend selectively to either the correctness of the grip or the goal of the observed action. Subjects were faster when attending to the goal of the action and a strong interference of goal-violations was observed when subjects attended to the grip of the action. Importantly, observation of irrelevant goal- or grip-related violations interfered with making decisions about the correctness of the relevant dimension only when the relevant dimension was correct. In contrast, in Experiment 2, when subjects attended to an action-irrelevant stimulus dimension (i.e. orientation of the object), no interference of goal- or grip-related violations was found, ruling out the possibility that interference-effects result from perceptual differences between stimuli. These findings suggest that understanding the correctness of an action selectively recruits specialized, but interacting networks, processing the correctness of goal- and grip-specific information during action observation.