The phrase "mental representation" in recent investigations of cognition is Janus-faced. On the one hand, it is used in discussions which, like many in past theorizing about the mind, are concerned with the nature of states such as beliefs and desires. The other face is turned toward contemporary cognitive neuroscience. The first perspective encompasses ordinary explanations of actions, emotions and beliefs; the second perspective invites an engagement with neuroscience.

Are the mental representations of the two discourses at all the same? The seldom-considered possibility that the two are not the same is in effect raised by an unjustly neglected paper by Stephen Stich (1992). Stich distinguishes between two families of projects: giving a philosophical account of ordinarily recognized intentional states, and giving an account of what is needed in cognitive science in the way of mental representations. He rejects the standard methodology of the first and criticizes recent philosophy for both failing to distinguish the two and failing to undertake the second.

In this paper, I investigate how "representation" actually is used in some areas in cognitive neuroscience. I will maintain that recent neuroscience should lead us to recognize two different genera of representation. Describing the features of an unrecognized genus of representation will be the major task of this paper. As we will see, the discussion points us toward an important conclusion: the actual domain of representation cannot be adequately described, let alone understood, in terms of our present theories of representation.

The arguments below invoke an important distinction. The distinction is one I will label as a distinction between the representational genus and the constitution of a representation. Though the discussion below will clarify the distinction, it will be useful to have an indication of it as we start.

There are very different kinds of representations; the representational genus a representation belongs to is the kind it is. The constitution of a representation is what about it makes it that kind of representation. Most recent literature on mental representations agrees fundamentally about the genus or nature question: There is, the literature suggests, one kind of representation and members of that kind are distinguished by having contents which give them satisfaction conditions. However, there is very considerable disagreement about qualifies something as a member of that genus. Recent theories have invoked such factors as evolutionary design (Millikan, 1989) and hyperspace vector analysis (Churchland, 1995 and 1998). Our focus will be on the question of whether there is more than one representation genus, and so we will only indirectly engage the disputes revolving around constitution questions.

Many recent philosophers have assumed that a theory of mental representation must have some sort of close relation to a theory of intentional content, either providing such a theory or elucidating some legitimate core of that notion. The notion itself has resisted precise analysis, but there are a number of related characteristics which are clustered around it. Something with intentional content is about something. Many recent philosophers would stress the fact that a representation may misrepresent. Representations have satisfaction conditions and these may or may not be met.

Recent philosophers speak of representations as though by definition they have intentional content and aboutness. For Robert Cummins (1996), the first problem for mental representations is the problem of content. He says, "The term 'representation' is just a generic semantic term" (p. 2). According to Fred Dretske (1995, p. 28) , "... [W]e find that a representational account of the mind provides a satisfying explanation of intentionality." Kathleen Akins's (1996) believers in the aboutness of representations include Daniel C. Dennett, David Papineau, Denis Stampe, and Kim Sterelny. To this list could be added William Lycan (1993), Christopher Peacocke (1994), and many others. Jerry Fodor (1976) is almost too obvious a proponent of the thesis to need mentioning. Indeed, the idea that there are inner representations with intentional contents can seem to be the defining distinction between cognitive neuroscience and behaviorist psychology (Richard Montgomery, 1995).

A recent manifestation of this assumption about representations is in Valerie Hardcastle’s (1999) The Myth of Pain. Hardcastle speaks interchangeably about representations, content and the intentional. The central elucidation we are given is in terms of "aboutness." Yet Hardcastle also takes representations so understood to be employed in neuroscience.

The examples of representation in the philosophical literature distinctly reflect the idea that representations have content. In these examples, gauges represent the pressure of tires (Dretske, 1995), vectors represent faces (Churchland, 1995), neural firings represent insects (Millikan, 1989), and pictures represent rural scenes (Crane, 1995). In the discussions provided of these examples, they all are thought to possess the aboutness and the capacity for misrepresentation that we saw above.

Many historians think that a theory of mental items with intentional content links current philosophy with its past. Thus, in a recent dispute otherwise full of disagreements, Robert Stecker (1992) and John Haldane (1993) agree that it is unproblematic to hold that the historical philosophers from Descartes through Hume held a Representational Theory of Mind where this theory is correctly defined by Jerry Fodor in terms of inner representations with intentional content. Each also agrees that theory of ideas was brought in to account for the intentionality of thought. Of course, neither is maintaining that all the 17^th and 18^th Century philosophers would have agreed with Fodor that the content must be language-like in its form; intentionality is, rather, to be understood basically in terms of aboutness and misrepresentability.

While there are well known difficulties in seeing connectionist networks as possessing discrete representations, many theorists assign representations to such nets of the sort just described. Nothing in the account of intentional representations requires that we be talking about the symbolic representations of classical models, at least as far as the literature goes. The alternative to locating content in such nets is, it is often thought, to eschew a realism about representations.

I shall argue that it is false that mental representations must have intentional content by locating a different conception of representation and demonstrating that it is in use in recent cognitive neuroscience. Since cognitive neuroscience has played a very considerable role in the recent conviction in philosophy and psychology that inner contentful representations are extremely important, it is very significant that whole stretches of cognitive neuroscience employ a notion of "representation" that is quite different from that in philosophy of mind.

Within the context of philosophy of mind, it may seem self-contradictory to say some X can represent Y but not be about Y. However, there are plenty of occasions outside of this restricted context that we can talk about non-intentional representing. For example, an elected official may represent a district without being about the distinct. Here clearly the marks of the intentional do not apply. While an official might be said to misrepresent the desires or opinions of the electorate in their votes, those of us who radically disagree with those who get elected know that we are nonetheless technically represented by them; if they do vote, we get committed by that vote to its being the case that our distinct voted that way. Their vote constitutes our vote and cannot misrepresent it, however much it more informally misrepresents our desires and interests. Further, while the representations featured in philosophy can represent a non-existent object, the official cannot so represent if there is no district to be represented. Perhaps elected officials are "representatives", not "representations." However, an elected official can be satisfy our "right of representation" and those of us with elected officials participate in a representational form of government.

The kind of representation we will focus on is distinct from either intentional representations or the representatives-as-proxies that we have just seen. To see some examples, we can start with representation questions. A "representation question" is an informal notion and indicates questions that could be answered correctly by giving what is generally counted as a representation in philosophy. Consider, then, the following questions:

• What kind of pet fish do you like?

• What color was the jacket you bought?

• What did she say that upset them so?

The following, let us suppose, are correct answers:

• Goldfish.

• It was the color of garnets.

• She swore three times.

These answers have what is typically called intentional content. The three answers are, respectively, about the (purported) facts that one particularly likes goldfish, that the garment is the color of a garnet, that she swore three times.

There is a different way to answer these questions. To the question of pets, if one is sitting in a friend's room containing goldfish in a bowl, one might simply say, "These." To the second question, one might simply point to another example of the color and perhaps say, "This." To the third question, one might respond, "Damn. Damn. Damn." In answering one has pointed to, or produced, an instance of the kind of thing in question.

I am going to call the instance pointed to or produced by the second set of answers a "token-realization," in contrast to the "intentional-representations" of the first sort of answer. At the end of this section, I will address the objection that token-realizations are covertly very like intentional-representations. For now, at least on the face of it, token-realizations are not about, e.g., goldfish or what one likes in the way of pets; they are of their type in the sense of being instances of their type, but not in a semantic sense of "of." That is, token-realizations are not of their type in the sense of meaning their type. (Denying this would seem to make intentional content possessed by everything that is an instance of a type and so to confuse intentional-representation and instantiation.) The chair I point to in giving an instance of the color of the garment does not itself mean anything about the garment. The token "Damn" is another token of the type her tokens were of; it is not itself about her previous tokens, or anyone else's.

Intentional-representations and token-realizations contrast regarding possible non-existence. These token-realizations cannot exist without, respectively, an instance of that color, fish, or instances of that word. If there are no goldfish, there can be no token-realizations of goldfish. Not so "Goldfish," which can certainly exist in a world in which there are no goldfish.

Token-realizations can have some advantages as answers to the representation questions; one may be able to perceive them and even interact with them. If a sighted interlocutor has no idea what the color of garnets is like, the sample may be much more informative than the phrase "like the color of garnets." Given the presence of the goldfish, one's hearers can observe the tokens and try to figure out what one likes about them, judge the wisdom of one's preferences, and so on.

A token-realization may realize a quite abstract type. Thus, items in a model of one's garden may realize the relative locations realized by the actual plants in one's garden but not realize the same distances at all. In general, the notion of isomorphism provides a sense of "realizing the same relation," which can apply at varying degrees of abstraction. (There are also well-known limits to what the notion of isomorphism can provide for the notion of a representation, and these will be discussed in the penultimate section of this paper.)

Allowing that token-realizations do represent is in accordance with the diversity of "represents" that the Oxford English Dictionary, among other things, recognizes:

1. Representing as presenting again: Something is present again.

2. Representing as standing in for: A parent represents a family at a ceremony; an elected official represents a constituency.

3. Representing as example: A murder might represent the kind of situation where conviction will be easy.

The only recognition of these diverse uses of "represents" that I have found in the philosophical literature is the semi-serious invocation of 2, the representatives as proxies. However, 2 is distinctly absent from the cognitive neuroscience literature, though 1 and 3 definitely are present. We can assimilate cases of 1, presenting again, to cases of 3, examples. With both 1 and 3, we have the idea that getting a representation is getting a token of a type. Using "token-realization" for both cases of 1 and cases of 3, we will end up with a gerry-mandering term; having such a term may be a consequence of the central tasks of this paper: both to develop a non-intentional account of representation and to link the resulting notion to an extensive use of "representation" in the cognitive neuroscience literature. Once this is done, it might be of value to posit different species under the non-intentional genus. Or it might not.

Accordingly, a token-representation, T, represents some individual, I, just in case either T is identical to I or there is some feature or kind of which T and I are both examples (i.e., coinstantiate). Thus, my "Damn" represents her utterance. Relatedly, T may represent a kind or type by being an instance of that kind or type; here is where the idea of an example in most at home among token-representations. For example, that fish’s behavior may represent the sort of fishy behavior one finds beautiful.

It is true that cases of 3, examples, are colloquially more often spoken of as "representative" rather than "representational." However, there are clear cases in recent cognitive neuroscience where researchers used "representations" of cases in which we have examples and not intentional contents. We will look at some of them in the following two sections.

Let us first ask if token-realizations are the covert bearers of intentional contents. Someone might claim that the tokens are really symbols with contents. They are in fact, it might be said, parts of the linguistic acts whose contents are about colors, goldfish and words

The central premise of this argument - that they are functioning as symbols — is wrong in important ways. First of all, it is wrong merely on the level of semantic theory. If I say of a cake, "This tastes good," I do not thereby make the cake into a symbol; it is, rather, merely the referent of a symbol. Similarly, if I say, "This is an example my favorite color," I do not make the referent of "this" into a symbol.

It may, of course, be that one's interlocutor needs more before the referent is clear; one may need to say "this color" or "this animal." What this fact highlights is the fact that the language is functioning to fix the reference. It falls far short of showing that the representation task is not being done by the referent.

What the comment does do, however, is remind us that many of our examples of intentional-representations depend on language-based interpersonal interchanges in that language (including the "language of gauges") creates the representation. This requires human deliberate behavior, and there are very serious questions about whether we can simply move notions of content which fit this sort of context into discourse about the brain. However, our token-realizations of colors and goldfish are language dependent in a very different way, for language merely fixes the reference to the representation; it does not create the representation. When we move from interpersonal communication to the brain, where we largely lack deliberate human interpersonal behavior, it may be that token-realizations will be theoretically much more plausible.

I will suggest eventually that there are at least two different central paradigms for the references to internal representations that occur in discussions of cognition. One source for the attribution of representations comes from the ascriptions of propositional attitudes. The other source is quite different; token-representations do not have satisfaction conditions and they do not have the propositional structure that at least many theories attribute to many of our internal states. As we will see eventually, we can be realists about much in cognitive neuroscience without being realists about the first sort of states. This is not to endorse an anti-realism about so-called folk psychological states; rather, it is to separate realism about such things from realism about representations in at least a wide expanse of neuroscience

The distinction between token-realizations and intentional-representations is not the same as that between semantically described cognitive states and purely syntactically described ones, or that between personal and sub-personal descriptions. One reason for this is that the taxonomy for token-realizations is not purely syntactical, but neither does it require the notion of representational content. In fact, the phenomenon of token-realizations suggest that the debates over a syntactic theory of the brain are misconceived, since the alternatives "pure syntactical" or "semantical" are not exclusive. In addition, token-realizations occur at many different levels of explanation for many different enterprises; they occur at sub-personal levels; we use them at the personal level to represent by quoting, and so on.

Token-realizations may represent an individual in one of two different ways. A token-realization may represent by being a token of the same type as that individual OR it may represent by realizing the individual. Token-realizations which realize are mentioned in the selections below:

While the conceptual sophistication present in good work in cognitive neuroscience is undoubted, philosophers may be disinclined to expect the same care with vocabulary. Hence, the following remarks may seem merely careless:

To understand the above remarks, suppose in the first instance we are talking about a pain and in the second instance about having an image of a cat. The first theorist, the text strongly suggests, does not think that there are two things, the internal representation and the pain. The second theorist does not think, the text again strongly suggests, that the neural representation of the image is distinct from the image. In such cases we have a realization of Y in X; in short, we have our token-realizations.

Another area in which a similar usage occurs is that of "knowledge representation." Here research is concerned with how knowledge is or can be realized. In the quote immediately following, "represent" could be replaced by "realize:"

The official philosophical story about cognitive neuroscience is that it employs an intentional-content conception of representation. However, something else is also going on. It should be noted that the researchers we will next look at themselves may not explicitly distinguish between representations as token-realizations and representations as bearers of intentional contents. The important point is that thinking of representations as token-realizations makes better sense of their discourse.

There are three specific areas with which we will be concerned. The first concerns cases of imagining and remembering, including imagining and remembering emotions and bodily actions. The second has to do with language and the third with perception. We will often focus our attention on specific articles from the scientific literature. The point to be stressed in each case is that the representations being spoken of are token-realizations, not bearers of intentional contents.

Area One: Imagining actions. In "The Mental Representation of Hand Movements after Parietal Cortex Damage," Sirigu et al. (1996) examine how one has knowledge of spatial and temporal features of one's actions before they are performed. Subjects are asked to imagine themselves making some motions. The authors talk, as the title indicates, about the resulting mental representations. These representations have a crucial feature:

While the article uses freely terms such as "images" and "representations," and thus invites interpretation in terms of the discourse of intentional contents, such representations are of the same neurological type as the neural parts of the actual actions. What we have, then, are something like brain routines which can be run on-line, and produce movement, or off-line and amount to imagined movement. The representations the authors speak of are not additional to the active routines; rather, they are token-realizations of these routines.

The authors in this article talk about simulation. The difference, as I have been emphasizing the evidence tells us, between a simulation-token and a real-token in these cases is not a difference in general neurological type. In addition, recent research indicates that just as we simulate actions by re-enacting parts of them, so in simulating emotions we undergo many of the same physical changes (Ledoux, 1996). The result is the same neurological type, as the perhaps too easy transition (for many of us) from remembering or imagining to the actual feeling indicates.

We will discuss perception in our third area of case studies, but there are important connections between the work we have just seen and recent work on perception. The perception of another’s action, at least in some species, may enable one to undertake the action oneself more easily. Underlying this is an important feature of the perception of action; namely, perceiving an action may involve forming the sort of neural representation of the action we have just seen Sirigu et al discuss, where the representation is in fact a token-realization of part of the action type. Thus, writing on the representation of action in rats, Leggio et al (2000) maintain,

What the authors term a "representation" is in fact a token-realization.

Area Two: Language. Bayes et al. (1998) are concerned with the brain bases of the language abilities of a split-brain patient; that is, someone whose brain hemispheres are independent after a resection of the corpus callosum. The article presents research that challenges the familiar idea that, as expressed elsewhere, "language abilities can be localized to one of the two cerebral hemispheres" across our species (Kandel et al (1995, p. 648). The authors argue that in some people the neural bases of spoken language may be in a different hemisphere from those of written language: "Spoken language may be innate, an "instinct" if you will, but writing does not develop without instruction. The pattern of lateralization observed here suggests it can have a more varied neural representation as well." (Here, as elsewhere in the discussion of this article, the emphasis is mine.)

The representations further appear to be of spoken and written language. They remark, "To better assess the neural representation of spoken and written language in V.J., we used lateralized testing techniques" (p. 902).

These authors clearly speak of representations of words and representations of language. Why interpret their descriptions of these representations as showing the presentations are thought of as examples of language? Why should their "neural representation of spoken … language" be thought of as a representation which represents by instantiating a type?

The authors, as is certainly typical of much discourse in the field, do place language, and the "language system," in the brain. As they say further (my emphasis):

The language system and language abilities are definitely said to be in the brain. The representations of words are, then, part of a model that realizes features of language in the brain. As a consequence, they are token-realizations.

The idea that there is a language in the brain -- and so token-realizations of language types in the brain -- arises early on in the development of the cognitive neuroscience literature. Its impact on the literature in cognitive neuroscience has been immense and it may indeed be a source of the easy acceptance of the idea that words are in the mind. However, there is more behind the idea that representations of words may themselves be words. We can represent words by using symbols that are "about" words and so have words as their intentional objects, but we also can represent words by using tokens of the same word-type OR tokens of another word-type, even one from a different language or code. Our initial example of uses of the "damn" is a case in point. In this latter case, what we have are token-representations.

The important point here is really an interpretative one: "Representation," used in the article above, is not about having intentional contents in the head that are about language; rather, it is used to signify instantiating whole or parts of language in the head. The representations are token-realizations. Of course, agreeing with my interpretation does not involve agreeing with the authors that the language system is in the brain; rather, the proposed interpretation in effect draws out a consequence of saying that the language system is in the brain. My point here is not with the truth of what is in the head, but rather with what is being said to be in the head.

We have looked at two sorts of case whose basic features are quite different. With each sort, the claim that the representations spoken of are actually token-realizations is argued for differently. In the first set of cases, the claim is that the represented and the representer are simply said to be (approximately) the same neurological type. Thus they directly met the conditions for being token-realizations. The argument in the second case depends on an claim that attributes a consequence of the authors' placing "the language system" in the brain. The claim is that a neural token that has word features is a token of such features. Given this is right, then we once again have token-realizations.

Area Three: Perception. That token-realizations play a large role in perception can be argued for in a still different way. The argument rests on two facts: (1) if a representing relation is constituted by isomorphism, then what we have are best interpreted as token-realizations (in the absence of objections yet to be developed); (2) current computational theories of perception in cognitive neuroscience just are theories that take a representation in the brain to be isomorphic to the perceived situation. That (2) is true is no longer a matter of articles in scientific journals; it is now both a textbook truism (Gazzaniga et al, 1998) that has even become a journalistic "fact"(Johnson, 2001). When a textbook asks, "What do neural signals in the visual pathway represent?", the answer is going to draw on the thesis that there is a set of algorithms to map features of the environment onto firing patterns in the brain. But such mapping just is a sufficient condition for isomorphism. Given the isomorphic result counts as a representation, then the antecedent of (1) is met.

While mapping is a sufficient condition for isomorphism, it may not be a sufficient condition for an interesting and effective theory of perception. We may need much more complicated relations than those suggested by a simple appeal to algorithms. This idea that we can adequately characterize what is needed by a comparison to literal maps is quite certainly faulty. It is, further, more than a mere logical possibility that the isomorphism constituting, for example, the representation of shape will turn out to be different from that constituting the representation of color. This is precisely what we should expect, given that what constitutes co-instantiation varies with what is being instantiated. Thus, while we can say that isomorphism can make something a token-realization, the term "isomorphism" has to be understood as a very general term which applies to a number of kinds of relations.

In this section, we have looked at three different sorts of area in which we located a different conception of representation operating in the neuroscientific literature. To claim that the representations instead function because they in addition have some intentional content is to attribute to them an extra feature, a content or an "aboutness" that has resisted both clarification and explanation. In the next section we will look briefly at two problems for token-realizations. First we need to consider an argument to say that the token-realizations in the neuroscience literature are really intentional-representations.

It might well be said that token-realizations can misrepresent, a major mark of intentional-representations, and so have intentional content. This is an important and potentially telling objection. I will address it in two steps. First of all, I will argue that nothing mandates the use of "true" or "correct" in such contexts, and then I will locate a different account of the right location for such assessments.

We should start by noting that users of a language can often use "true" or "correct" inappropriately. For example, if an elector from Florida in 2001 were to have said, to cast his vote in the electoral college, "My state votes for Bush," many people might well have said what he said was false. This does NOT show that the representative is correctly understood as attempting to describe the way votes were actually cast in Florida.

Do the neural events possess satisfaction conditions? Are they true or false? The details of the answer to the question may vary greatly with the sort of case we are considering. At the same time, the overall conclusion of this paper needs really only a limited number of cases. It is enough for now if we can establish a good case for saying there are some uses of "representation" in the neuroscience literature which do not employ an intentional-representational conception. So I will focus on perception, and one kind of case. I think, however, think that the general pattern of response is applicable in all the sorts of case we have considered so far.

Suppose the spiking patterns in different people’s neural reactions to an environment are different in a particular respect: While A’s evoked potentials are equal in amplitude for two instances of sound of equal magnitude, B’s are not, and the amplitude of B’s decreases with the second sound. Does this mean that B is misrepresenting the sounds as different in magnitude? Well, we can say so, but should we?

The difference between A and B is classically exemplified in the difference between schizophrenic and non-schizophrenic subjects in two-click tests of auditory gating. Schizophrenics appear not to have the capacity to "tune out" that non-schizophrenics have; thus their spiking patterns for the first and second clicks tend to remain the same. Nothing at this level shows that one spiking pattern is true and the other false. Both, however, are said to represent the sounds in the environment. What this shows is that the way representing is functioning here does not require that the representations are assessed as true or false.

This phenomenon of auditory gating, and the similar visual masking, can lead to misperceptions and mistaken beliefs. In such cases, there are descriptions attributing to the subjects states that can be straightforwardly assessed as true or false. However, nothing demands that the truth values thus located should be transferred down to initial excitations in the nervous system.

Suppose one maintains that while the token-representations are not true or false, they are nonetheless correct or incorrect. We can locate a good context where assessments of "correct" and "incorrect" function very well with neural representations of the sort we have looked at, and these are quite different from the typical construal in philosophy of matching up with the world. The most important dimension for neural representations of the sort we have been looking at is their ability to mesh with other mechanisms in a way which enables us to move effectively through our environment. Thus, the neural mechanism that registers two identical clicks identically can be faulty because it impedes effective behavior; noise become counterproductively distracting. This dimension — the usefulness of representations, and not their "correspondence" to the world — is the dimension of assessment most appropriate and important. That discussions of the traditional notion of intention-representation have so stressed getting the facts right is a dramatic difference between representations in philosophy and representations in neuroscience.

In conclusion, we can note that one disutility token-representations can have is that they can be part of what leads to false beliefs. Thus, someone whose gating capacities are intact may not notice the second click, or may think it is less loud. On the other hand, that same person may notice the click and may be able to judge it correctly. Hence, the fact that the token-representation MAY lead to false beliefs is not a good reason for saying that it will. Further, given that the error comes in at a higher level, it seems wrong to locate the error in the token-representation.

Of course, neuroscience is not without an interest in how we do get the facts right and cognitive neuroscience textbooks these days use "representation" in many different contexts, including some where questions of truth are certainly necessary. My point has been a modest one: there is a range of uses of "representation" and "represents" in neuroscience which do not fit the philosopher’s picture. That discourse points us toward a very different kind of representation.

Accounts of mental representations have tended to assume that there is one kind of mental representation, intention-representation. Such accounts are then saddled with the task of developing a theory of meaning and content which is supposed to provide a unified account of the truth-conditions for our ascriptions of representations to organisms. That is, there is supposed to be a single story underlying "The cat senses danger," "The frog sees a fly," and "Susan added two and two." The story is the story of how inner states acquire intentional content and of how possession of inner content entails the truth of attributions of contentful states to the organism. One kind of story invokes the way cognitive mechanisms are supposedly designed to promote the survival of the organism; this sort of story has been developed by Millikan and Dretske, among others. Another kind of story focuses on causal origins of the representation and attempts to find a kind of complicated causal relation that will discriminate between content-providing causes and others; Fodor's work is prominent here (Fodor, 1990). Cummins attempts what might be counted as a third sort of theory, though it draws on causal-teleological elements. Yet other accounts may focus on the causal roles the representation plays among other representations. And all these accounts may also borrow elements of yet another kind of story, one that refuses to locate content principally in the head; rather, the content attributed to organism such as ourselves at least supervenes on, among other things, social practices.

In all these theories, the question of the nature of the representation has gotten some agreement; representations have a content that gets the facts right or wrong; hence, they represent or misrepresent. The constitution question is the one very much in dispute, as the variety of theories about what the constitution amounts to shows. The very fact of the dispute additionally demonstrates the difficulty of providing a theoretical account of intentionality and the accompanying problem of misrepresentation.

Token-realizations provide an ontologically much more modest commitment. Rather than introducing a problematic new element, intentionality, into the world, by definition they merely use features already there. A token-realization is merely an instance. This virtue may, however, seem quickly to give way to two at least seemingly undesirable features.

The first undesirable feature is symmetry and the second is an extraordinary promiscuity (compare Eagan, 1998). If A is a token-realization and as such represents B, then A and B coinstantiate some property. However, B will also then represent A. Hence, "represents" is symmetrical; if A represents B, then B represents A. But this does not seem to accord with the way in which "represents" is used in neuroscience, and elsewhere. Further, A will represent anything with which it co-instantiates some property, and if A has mass, it shares this property with all things with mass, and so on. But a notion according to which anything represents just about everything is useless.

There are three strategies for dealing with such problems. One strategy is just to embrace the features as features of a different sense of representation, and to declare that the problem really resides with those who have helped themselves to a more problematic notion of representation, an opaque notion for which they have no adequate account. There are two more conciliatory strategies: we can attempt to modify the semantics or we can add in the pragmatics. There are two good reasons for seeking a solution through the pragmatics. First of all, the notion of being an example, which provides much of the informal understanding of "token-realization," is a semantically similarly wildly promiscuous notion, but it is also much more useful than the promiscuity would suggest. We talk about examples and cite examples a very great deal. Hence, there’s a clear example where a notion is much more useful than its semantics alone can guarantee.

Secondly, "representation" figures in neuroscience as part of an explanatory enterprise and, more generally, as an element of actual causal explanations. We already have plausible accounts which appeal to pragmatic considerations to explain why something gets cited as, e.g., "the" explanation. At least in terms of the literature, it is plausible to think that pragmatics shapes some quite deeply embedded features of explanations.

What sorts of pragmatic considerations shape representational explanations to provide the seeming asymmetry and the non-promiscuity? There are clear goals in representation explanations in cognitive neuroscience. The goals are not just to describe in some sense ‘how the nervous system works;’ rather, features of the nervous system are generally being positioned as effects of the external world and they are to be described as potentially affecting further systems in order to produce, when the organism is functioning well, effective behavior. In effect, we have two factors that appear, singly or together, in standard accounts in the literature of what constitutes an intentional-representation: causality and how the system’s correct functioning promotes the organism’s survival. If these features are sufficient to provide a semantics that gives us asymmetry and non-promiscuity, then they will certainly be enough to give us a pragmatics that does the same. They would even indeed explain why it may to some seem permissible to talk about spiking patterns as true or false (though, perhaps I should say, this is not something to which I feel the least temptation). Thus, token-realizations are at least no worse off than intentional-representations, for whatever is employed in answering the constitution question can create the needed pragmatic features of token-realizations.

Notice that talking about taking features of the nervous system and its reactions as promoting survival is not necessarily to see them as literally designed by evolution; rather, it could be merely a matter of our taking a design stance. Of course, appeals to design stances, as opposed to more literal appeals to evolutionary design, will not yield the intentional realism aimed for by theorists such as Millikan, Dretske and Cummins. However, we can be realists about token-realizations without ascribing to them intentionality at all.

In this section, we have looked at some problems for any account of token-realizations. Pragmatic considerations offer us a way to deal with them. The presence of pragmatic considerations to secure some of the features that intentional-representations seem naturally to possess raises an interesting question: When do we need intentional-representations and when do we need mere token-realizations? If pragmatic considerations can shape our discourse to make it look more like discourse employing intention-representations, how do we tell which is which?

I suggest that the question just asked should make us pause at philosophical discussions today about representations. If there really is, as I have been arguing, a robust notion of representation that does not require any notion of intentionality, it may well be that there are two very different models of representation operating in the discourse employing "represents" and "representation." Language/code use and language content may provide us with one, and examples or token-realizations may provide us with another. If this is so, the general domain of representations has yet to be given an adequate description. The possibility that there is an indeterminate middle between two such paradigms cannot even be articulated within the outstanding theories of the nature of representation today.

Suppose one says, "Olympia Snow is a US Senator." How and what might one’s utterance represent? One way is the way familiar in philosophy. The utterance is about Olympia Snow and it claims she holds a particular office. What we have seen in this paper is a different way it might represent. For example, the utterance might represent what someone else had just said, by being an utterance of the same type as the earlier utterance. Similarly, the beginning of this very paragraph might represent to its readers the sort of sentence that shows up in philosophy papers still influenced, albeit perhaps unintentionally, by ordinary language philosophy.

When scientific theorists talk about images or pains being represented in the brain, they may seem to philosophical ears to be mistaken in their terminology. I think it is the philosophers who are mistaken here. There is another conception of representation which the scientists are employing. It is a very important notion, which promises access to a notion of representation that is free of the host of exceptionally difficult problems carried by the conception of intentionality that philosophy has for too long thought to be essential to any notion of representation.