In 1997, Jochen Müsseler and Bernhard Hommel asked participants to prepare a left or right keypress and then identify a briefly flashed arrow. When the arrow pointed in the same direction as the prepared action, people were worse at identifying it. Planning a movement had reached into perception and changed what participants could report seeing.

Findings like this sit awkwardly with a long-standing picture of the mind. Susan Hurley called it the classical sandwich: perception delivers representations, central cognition reasons over them, and the motor system executes the results. On this model, action is output. It shouldn't shape input, and it certainly shouldn't take part in thinking itself.

Yet a growing body of evidence suggests that motor systems do far more than move limbs. They appear to structure perception, support our understanding of other people's behaviour, and may even scaffold abstract concepts. The interesting philosophical question is not whether action matters to cognition, but how deeply it matters, and what that reveals about the architecture of mind.

Wolfgang Prinz's common coding theory offers a useful starting point. Prinz proposed that actions are represented in terms of their perceivable effects, so planning a movement and perceiving an event draw on a shared representational format. Hommel and colleagues later formalised this as the Theory of Event Coding. The prediction is straightforward. If perception and action share codes, each should be able to facilitate or interfere with the other, and in both directions.

The influence of perception on action is well documented. In Tucker and Ellis's 1998 study, participants judged whether pictured household objects were upright or inverted. Handle orientation was irrelevant to the task, yet responses were faster when the responding hand matched the side of the handle. Merely seeing a graspable object appears to partially prepare the motor programme for grasping it. This is an automatic activation of what Gibson would have called the object's affordances.

The reverse direction is more surprising. Beyond Müsseler and Hommel's action-effect blindness, Hecht, Vogt and Prinz found that people trained to produce rhythmic arm movements without visual feedback later became better at visually discriminating those same timing patterns. Motor learning transferred to perceptual skill, even though participants had no perceptual practice. Whatever they acquired seems to have been stored in a format that both systems could read and use.

Caution is warranted, however. Firestone and Scholl have argued that many headline claims, such as hills looking steeper to people wearing heavy backpacks, reflect judgement or response biases rather than genuine changes to perception. The more robust findings concern interactions at the level of planning and event representation, not a wholesale reshaping of early vision. That distinction matters philosophically. It suggests that Fodor's modular input systems may be more permeable at their outputs than at their core.

Takeaway

Perception and action may not be separate stages in a pipeline but two ways of drawing on a shared representational currency. The open question is where that sharing begins and where it ends.

The modern story begins in Parma. In 1992, di Pellegrino and colleagues reported neurons in macaque premotor area F5 that fired both when a monkey grasped an object and when it watched an experimenter perform the same grasp. Umiltà and colleagues later showed that some of these mirror neurons responded even when the final stage of a grasp was hidden behind a screen. That result suggests they encode action goals rather than visual patterns alone.

Human evidence followed. Fadiga and colleagues used transcranial magnetic stimulation to show that watching someone grasp an object increases motor-evoked potentials in the observer's corresponding hand muscles. Calvo-Merino and colleagues scanned ballet dancers and capoeira practitioners as they watched both styles. Premotor and parietal regions responded more strongly when participants viewed the style they had trained in. Expertise in doing appears to shape the brain's response to seeing.

Aglioti and colleagues pushed this further with elite basketball players. Shown video clips of free throws cut off before the ball left the shooter's hands, players predicted the outcome earlier and more accurately than coaches or sports journalists with comparable visual experience. Players also showed outcome-specific changes in corticospinal excitability. The motor repertoire seemed to supply predictive information that visual familiarity alone did not provide.

These findings bear directly on folk psychology, lending support to simulation theory over pure theory-theory accounts of how we understand others. Yet Gregory Hickok and others note that patients with motor damage often still understand observed actions. Motor activation might accompany understanding rather than constitute it. The more defensible conclusion is narrower: motor simulation likely contributes fine-grained predictive models of how actions unfold, without being the whole of action understanding.

Takeaway

Your own motor repertoire may work less like a dictionary for decoding others' actions and more like a forecasting engine, one that is sharpest for the things you have actually learned to do.

Fodor's language of thought treats concepts as amodal symbols, manipulated according to their syntax. Lawrence Barsalou's perceptual symbol systems theory proposes something different. On his view, concepts are partial reactivations of the sensory and motor states active when we experienced their referents. This would address Stevan Harnad's symbol grounding problem, since meaning would be anchored in bodily experience rather than in further symbols.

Some evidence fits this picture well. Hauk, Johnsrude and Pulvermüller found that reading verbs like lick, pick and kick activated motor regions somatotopically, near areas controlling the face, arm and leg respectively. Glenberg and Kaschak reported an action-sentence compatibility effect, in which judging sentences was faster when the response movement matched the direction they implied. Strikingly, the effect seemed to extend to abstract transfer, as in telling someone a story.

The extension to abstract concepts draws on Lakoff and Johnson's conceptual metaphor theory: we grasp ideas and move through time. But the evidence has grown more contested. A large preregistered multi-lab replication of the action-sentence compatibility effect, published by Morey and colleagues in 2022, found no reliable effect. Mahon and Caramazza have argued that motor activation may simply be activation spreading from amodal concepts rather than their foundation.

Many researchers now favour hybrid views. Anna Borghi's Words as Social Tools proposal suggests that concrete concepts lean heavily on sensorimotor grounding, while abstract concepts rely more on linguistic and social information. Grounding, on this view, comes in degrees. The productive questions become more precise: which concepts, in which tasks, and does motor involvement play a constitutive role in meaning or only a contextual one in processing?

Takeaway

The better question may not be whether concepts are grounded in action, but how much grounding each concept needs to do its work, and how much of that load language can carry instead.

Across perception, social understanding and conceptual processing, the motor system turns up where classical architectures said it shouldn't. That alone is enough to unsettle the tidy sandwich of input, cognition and output.

Yet the empirical record also counsels restraint. Several striking effects have weakened under scrutiny, and patient data show that motor involvement is often neither necessary nor sufficient for understanding. Strong embodiment claims outrun the evidence.

What emerges instead is a more precise research programme. The useful questions now concern format, timing and necessity. Here cognitive science does what philosophy alone cannot: it turns broad debates about embodiment into testable claims about mechanisms, and gradually shows us where the boundaries of action-based cognition actually lie.