Disentangling the contributions of grasp and action representations in the recognition of manipulable objects