Loading field map
Mental representation is the study of how the mind carries information about the world. The core idea is that thoughts, perceptions, memories, and intentions are not the things they are about, but stand in for them. When you think about your childhood home, the thought is not the house; it is something in your mind that represents the house. The subfield asks what these stand-ins are made of, how they get their content, and how they can be about things at all.
The central puzzle is intentionality: the "aboutness" of mental states. A belief is about Paris, a desire is for coffee, a perception is of a red apple. This aboutness is remarkable because a mental state can be about something that does not exist (a unicorn), something that is not present (a distant friend), or something that may never happen (a future vacation). Mental representation is the attempt to explain this aboutness without treating the mind as a mysterious substance. It sits at the intersection of philosophy of mind, cognitive science, and linguistics, and it is the conceptual foundation for much of cognitive psychology and artificial intelligence.
Before any theory of representation can be built, philosophers must decide what a representation is. The most basic distinction is between the vehicle and the content. The vehicle is the physical or functional thing that does the representing—a pattern of neural firing, a sentence in a "language of thought," a mental image. The content is what the vehicle represents—the proposition, object, or state of affairs it is about. A theory of mental representation must explain how a particular vehicle gets a particular content, and why that content is what it is rather than something else.
This is known as the problem of intentional content, or the "aboutness" problem. It has several sub-problems. The first is the problem of error: a representation can be false or inaccurate. A belief that it is raining can be wrong. Any theory of content must explain how a representation can be about something and yet misrepresent it. The second is the problem of normativity: representations have correctness conditions. A perception is not just a pattern of activity; it is supposed to match the world. The third is the problem of causation: mental states cause behavior. If a belief is just a pattern of neurons, how does its content—what it is about—make a difference to what the body does? These problems define the field's agenda.
The earliest systematic accounts of mental representation came from the empiricist tradition, particularly in the work of John Locke and David Hume. They argued that the mind's contents are ideas—roughly, mental images or copies of sensory experience. To represent a tree is to have an image of a tree in the mind. This view was intuitive and powerful: it explained how thought gets its content from experience, and it made representation a matter of resemblance. The image of a tree resembles the tree, and that resemblance is what makes it about the tree.
This "imagist" theory ran into serious difficulties. First, it could not explain abstract concepts. What image represents "justice" or "infinity"? Second, it could not explain the systematicity of thought. You can think "the cat is on the mat" and "the mat is under the cat" using the same elements, but images do not obviously combine into complex thoughts with a determinate structure. Third, resemblance is not sufficient for aboutness. A photograph of a tree resembles the tree, but it is not a thought about the tree; it is just a physical object. Something more than resemblance is needed.
The imagist view was not a school in the modern sense but a broad tradition. It was the default assumption for centuries, and it still has echoes in theories that treat mental imagery as a central form of representation. But by the late nineteenth century, philosophers such as Gottlob Frege and Edmund Husserl had begun to argue that mental content cannot be reduced to images. Frege distinguished the sense of an expression from the mental image associated with it. The sense is a public, shareable content; the image is private and idiosyncratic. This distinction set the stage for the modern problem: what kind of thing is a content, and how does a mind grasp it?
The most influential modern framework for mental representation is the computational theory of mind, often associated with the philosopher Jerry Fodor. Fodor argued that mental representations are not images but symbols in a "language of thought"—sometimes called "Mentalese." This language has a combinatorial syntax and semantics: complex representations are built from simple ones according to rules, and the content of a complex representation is a function of the contents of its parts and their arrangement. Just as the English sentence "the cat is on the mat" is composed of words with meanings and syntactic rules, a thought is composed of mental symbols with contents and combinatorial rules.
This view has several advantages. It explains the systematicity and productivity of thought: because representations have a combinatorial structure, you can think an infinite number of thoughts from a finite stock of concepts. It also fits naturally with the computational theory of mind, which holds that thinking is a kind of computation—the manipulation of symbols according to rules. On this view, a mental representation is a token in a computational system, and its content is determined by its role in that system.
The language of thought hypothesis is not a single theory but a research programme. It has been developed in different ways by Fodor, by the cognitive scientist Zenon Pylyshyn, and by many others. Its central commitment is that mental representations have a syntax—a structure that can be manipulated by computational processes. This is what makes it possible for a physical system, like a brain, to implement thought. The brain does not need to contain little pictures or sentences; it needs to contain patterns of activity that can be transformed according to rules that respect the structure of the representations.
The computational approach has been enormously influential, but it faces a major challenge: the problem of grounding. If mental symbols get their content from their relations to other symbols, how does any symbol get connected to the world? A purely syntactic system—a "symbol crunching" machine—could manipulate symbols without knowing what they mean. This is the "symbol grounding problem," raised by the cognitive scientist Stevan Harnad and others. A theory of mental representation must explain how symbols get their semantic content, not just their syntactic role.
One family of answers to the grounding problem is the causal or informational theory of content. The basic idea is that a mental representation represents what it is reliably caused by. A mental symbol for "dog" represents dogs because, in normal circumstances, dogs cause the tokening of that symbol. The representation is a kind of internal indicator or detector: it carries information about the presence of dogs, just as a smoke detector carries information about smoke.
This view has a natural home in cognitive science, where perception is often understood as information processing. The retina, for example, carries information about light patterns; the brain processes that information to produce representations of objects. The causal theory explains how content can be naturalistic: it does not appeal to mysterious mental powers, but to ordinary causal relations between the world and the nervous system.
The problem with the causal theory is error. If a mental symbol is caused by dogs, what happens when a wolf causes the "dog" symbol? The representation is false, but the causal theory seems to say that it represents whatever caused it—in this case, the wolf. This is the "disjunction problem," first articulated by Fodor. If the symbol is caused by both dogs and wolves, then its content is disjunctive: it represents "dog-or-wolf." But that makes error impossible, because the representation is always true of whatever caused it. A theory of content must explain how a representation can be supposed to be caused by one thing but actually caused by another.
Fodor's own solution was the "asymmetric dependence" theory. Roughly, the idea is that the symbol represents dogs because the causal connection between dogs and the symbol is more basic than the connection between wolves and the symbol. The wolf-to-symbol connection depends on the dog-to-symbol connection (because wolves are similar to dogs, they trigger the same mechanism), but not vice versa. This asymmetry is supposed to explain why the symbol is about dogs even when it is caused by wolves. This theory is ingenious but has been widely criticized; it is not clear that the required asymmetry can be specified without circularity.
Other versions of the informational approach appeal to teleology. The teleological theory, associated with Ruth Millikan and David Papineau, says that a representation has the content it was designed to carry. A heart is for pumping blood because that is what it was selected to do; similarly, a mental representation is for indicating a certain state of affairs because that is what it was selected for by evolution. The "dog" symbol represents dogs because the mechanism that produces it was selected for its ability to indicate dogs. Error occurs when the mechanism misfires—when it indicates a wolf in circumstances where it was designed to indicate dogs.
Teleological theories have the advantage of explaining normativity: representations have proper functions, and error is a failure to perform that function. But they face the problem of explaining how natural selection can determine content. Evolution selects for behavior, not for content; it is not clear that a mechanism has a single "proper" content rather than a range of contents that all contributed to fitness. The teleological theory also struggles with novel or learned representations, which are not the product of natural selection.
A different family of approaches rejects the idea that content is determined by causal relations to the world. Instead, content is determined by the role a representation plays in the overall system of thought. This is the "conceptual role" or "functional role" theory. The idea is that a representation's content is its place in a network of inferences and behavioral dispositions. The concept "dog" is what it is because of its relations to other concepts: it implies "animal," it is compatible with "pet," it is incompatible with "cat," and so on. To have the concept is to be disposed to make these inferences.
This view is attractive because it explains how content can be holistic and how concepts can be learned. It also avoids the disjunction problem, because content is not determined by what causes the representation but by how it is used. But it faces a serious objection: if content is determined by the entire network of inferences, then no two people have exactly the same concepts, because no two people have exactly the same network. This makes communication and shared content mysterious. It also seems to make content too "internal"—it cannot explain how a representation is about the world, only how it relates to other representations.
A more recent structural approach comes from predictive processing, a framework in cognitive science that treats the brain as a prediction machine. On this view, the brain does not passively receive information from the world; it actively generates predictions about sensory input and updates those predictions based on error signals. Mental representations are the predictions themselves—they are models of the world that are tested against sensory evidence. This view has been developed by Karl Friston, Andy Clark, and Jakob Hohwy, among others. It is not a single theory of content but a framework for understanding how representations are generated and revised. It has been influential in neuroscience and cognitive science, but its philosophical implications for mental representation are still being worked out.
A major challenge to all internalist theories of representation comes from externalism. The externalist argues that the content of a mental representation is not determined solely by what is inside the head. The classic thought experiment is Hilary Putnam's "Twin Earth": imagine a planet exactly like Earth, except that the substance called "water" has a different chemical composition. When an Earthling and a Twin Earthling both think "water is wet," their thoughts have different contents, even though their brains are identical. The difference is in the world, not in the head. This suggests that content depends on the environment, not just on internal states.
Externalism has been developed in several directions. One is the "extended mind" thesis, associated with Andy Clark and David Chalmers. They argue that mental representation is not confined to the brain. A notebook, a smartphone, or a calculator can serve as a part of a cognitive system, storing and processing information in ways that are functionally equivalent to internal memory. If a person with Alzheimer's uses a notebook to remember appointments, the notebook is not just a tool; it is part of the person's cognitive apparatus. This view blurs the boundary between mind and world and has implications for how we understand representation: representations can be external, distributed, and socially shared.
Externalism is not a single theory of content but a constraint on any such theory. It says that content is not determined by the intrinsic properties of the representing vehicle. This is compatible with causal and teleological theories, which already appeal to the environment, but it is a problem for pure conceptual role theories. It also raises the question of whether representation is a natural kind or a useful fiction. If content depends on the environment, then the same internal state can have different contents in different contexts, which makes it hard to study representation in isolation from the world.
The field of mental representation is not organized into a single dominant school. Instead, it is a set of overlapping research programmes, each addressing different aspects of the problem. The computational and language-of-thought tradition remains influential, especially in cognitive science and artificial intelligence, but it has been modified by connectionist and neural network approaches, which reject the idea that representations are symbolic and instead treat them as distributed patterns of activation across a network. Connectionist representations do not have a clear syntax; their content is spread across the network and is not easily decomposable into parts. This has led to a debate about whether connectionist networks can genuinely represent the world or whether they are just sophisticated pattern matchers.
The causal and teleological theories remain the main naturalistic candidates for explaining content, but neither has achieved consensus. The debate between them is active, and there are hybrid theories that combine elements of both. The conceptual role approach has been less prominent in recent years but has been revived in some forms of inferentialism, associated with Robert Brandom, which treats content as a matter of inferential commitments.
The most significant recent development is the rise of "4E" cognition—embodied, embedded, enacted, and extended. This movement challenges the assumption that representation is the primary mode of cognition. Enactivist philosophers, such as Alva Noë and Shaun Gallagher, argue that cognition is not a matter of representing the world but of engaging with it. Perception is not the construction of an internal model but a skillful exploration of the environment. On this view, representation is not the fundamental unit of thought; it is a special-purpose tool that some cognitive processes use, not a universal feature of mind.
This has led to a split in the field. Some philosophers and cognitive scientists continue to work on the traditional problem of representation, trying to naturalize content. Others argue that the whole framework is mistaken and that we should replace it with a non-representational account of cognition. This is not a settled debate; it is an ongoing disagreement about the very nature of the mind.
What remains durable is the problem itself. Whether or not the mind is fundamentally representational, it is undeniable that humans and other animals can think about things that are not present, plan for the future, and communicate about the world. Mental representation is the attempt to understand how this is possible. The field is characterized by a productive tension between naturalistic ambitions—to explain representation in physical terms—and the apparent irreducibility of aboutness. No theory has fully resolved this tension, and the field continues to be defined by the attempt to do so.