1. The Representation Objection

Jane rests her hand on a mossy boulder. She feels the damp moss and the outline of the boulder. A moment later she thinks the boulder obstructs my view of the waterfall. Both mental states represent things outside of Jane herself. But they differ in important ways. States of the first kind do not have a propositional structure; as a result, they lack a truth value. States of the second kind have ‘intentional content’ in the sense that they can be true or false. To avoid confusion, I shall reserve ‘representation’ for states of the first, nonpropositional sort. Both representations and intentional contents can be said to ‘be about’ or ‘represent’ something, namely, properties or objects in the first case and states of affairs in the second.

These two states of Jane—her feeling the mossy boulder and thinking that it obscures her view—are our explananda. We must distinguish two questions: (1) What is it that imbues a state with aboutness? (2) What is it that makes this mental state about this or that object or state of affairs? The two questions need not get the same answer. Even if one finds phenomenal character a plausible answer to (1), that does not commit one to giving the same answer to (2). The representation objection, as I'll call it, claims that PI so far has no good answer to (2).

Before developing that objection, we need a clearer notion of PI. First, PI claims that at least some intentionality is grounded in phenomenology, where this means, at a minimum, that it is in virtue of having such-and-such a phenomenology that a state represents what it does (Horgan and Tienson Reference Horgan, Tienson and Chalmers2002: 520). Some (e.g., David Pitt Reference Pitt2004) will go further and explain this grounding relation by means of identity: there is nothing more to the representational content of a state over and above its phenomenology. Moreover, phenomenal intentionality is basic in the sense that it cannot be derived from some other form of intentionality.

Another distinctive feature is the narrowness of phenomenal intentionality: it does not depend for its nature on anything outside the subject. Phenomenal intentionality is supposed to be exhaustively accessible to the subject undergoing it. A brain in a vat and its embodied counterpart will enjoy the same phenomenology and hence the same representational and intentional states (here I am ignoring a variety of complications that surround the internalism/externalism debates; for a careful disentangling of these issues, see Gertler [Reference Gertler2013]).

Within the PI camp, there are, of course, important divisions. Horgan and Tienson's (Reference Horgan, Tienson and Chalmers2002) view encompasses all of cognition. As they see matters, there is a distinctive phenomenology associated with entertaining any proposition whatsoever. ‘What it's like’ to think 2 + 2 = 4 explains why one's thought is about 2 + 2 = 4. But not all views endorse what we might call cognitive phenomenology (for criticism, see Pautz Reference Pautz and Kriegel2013). One might want to claim instead that the fundamental source of intentionality is perceptual. Perhaps aboutness enters the world when sentient subjects begin experiencing their environments, and more sophisticated cognitive states inherit their aboutness from such experiences. Since both views will maintain that there is at least perceptual phenomenal intentionality, I'll focus on this overlap region.

For our purposes, the core of PI is the claim that the aboutness of a mental state is not a function of its selectional or causal history but solely of its own nature. In contrast to the external relations the naturalizing program explored—call them ‘tracking’ relations—what makes a state about its object is, on PI, a nonrelational property. To use an example of Kriegel's: suppose you ‘undergo an experience with a squarish phenomenal character’ (Reference Kriegel and Kriegel2013b: 7). Simply in virtue of being in that state, PI tells us, you are thinking about a square object. We can say that aboutness is an external relation for tracking theories but an internal relation for PI.

As I use the terms, an internal relation is one that obtains solely in virtue of the intrinsic properties of the relata while an external one does not. Mass is (modulo some problems about the speed of light) an intrinsic property, weight an extrinsic one; having more mass than is an internal relation, weighing more than an external one. What makes for an intrinsic property is a subject of debate; see, e.g., Langton and Lewis (Reference Langton and Lewis1998). But the intuitive notion of a property an object has regardless of what is going on outside of that object is clear enough for our purposes. The decisive innovation of PI is its shift from external relations to internal relations. Consider, for example, David Pitt's claim:

Here are Graham, Horgan, and Tienson:

There are two closely related points here we must pry apart. First is the claim that the subject can ‘just tell’ what she is thinking. Jane does not need to consult a neuroscientist to know that she is thinking that the boulder is obstructing her view of the waterfall. That thought wears its character on its face. This first point is largely uncontroversial, at least if we bracket questions about wide content. In this sense, aboutness is not a relation at all; the nature of the state makes it the case that it has the content that it does.

But of course, this is only half the story. Sometimes, there are states of affairs (or properties) that intentional contents (or their constituent representations) stand in some relation to. And that relation, whatever it is, is internal in the sense that it obtains, when and if it does, solely in virtue of the intrinsic properties of the relata.

This second claim is far from uncontroversial. In fact, it should seem prima facie implausible, a return to the sort of picture Hilary Putnam (Reference Putnam1981: 3) derides as ‘magical’. For the second point is not a claim about first-person accessibility; it is a substantive claim about representation.

Nor is it optional. If phenomenology is to fix the reference of intentional states, then the relation between the phenomenology and whatever it is about must be an internal one. An external relation, such those in the tracking family, would deprive phenomenology of its alleged explanatory power. After all, nearly any state of anything can in principle track any other. By contrast, PI claims that it is the nature of phenomenal character and the states that have it that explains aboutness. Moreover, an external relation would presumably block subjects from using introspection to discern the narrow contents of their own thoughts.

PI is committed, then, to the claim that representation is an internal relation. But what relation, exactly? As far as I can tell, the proponents of PI have not told us. Let me distinguish this ‘representation objection’ from a closely related one. Adam Pautz argues that adherents of PI ‘tend to ignore the whole hard problem of naturalizing the mind’ (Reference Pautz and Kriegel2013: 195). Although I agree—and will do the same—that is not the issue here. Instead, the objection is that PI does not so far have a substantive explanation of representation. That explanation need not be naturalistic, whatever one precisely means by that. But without some story to tell, PI is significantly less attractive than it at first seems.

Perhaps proponents of PI see no need to offer such a story. Here is one line of thought they might have in mind. Suppose that the only kind of aboutness is propositional and hence that we need only explain how intentional content is tied to states of affairs. If so, there is no mystery about this tie: a state whose phenomenal character can be expressed as there is a boulder under my hand represents that state of affairs just in case there is a boulder under my hand. In other words, representation is truth.

The chief problem with this maneuver is that it fails to illuminate intentional content. Even those who agree that ‘what it's like’ somehow grounds ‘what it's of’ might want to be told how that grounding is accomplished. But there are a host of other problems as well. None is a knock-down argument against the representation-as-truth maneuver, but each adds to the costs that would have to be paid.

First, suppose my mental state presents the rock as being covered in moss when it isn't. Is that a false thought about a moss-less rock or a thought about a mossy rock that lacks a truth value because its putative subject—the mossy rock—fails to exist? The best answer is the first, but it is unclear how the present proposal is entitled to it. If the state's intentional content is false, there seem to be no grounds for saying that it is about or represents the state of affairs that makes that content false.

The natural reply is that the thought is about the rock in question. It just falsely predicates mossy of the rock. But the proposal under review has no way of tying subpropositional elements like rock to their objects in the world. Again, the proposal has it that representing works at the propositional level. The relata then are not simple mental items like rock on the one hand and rocks on the other; they must be propositional contents such as the rock is mossy and the states of affairs, if such there be, that answer to them.

Second, if thoughts such as the boulder is mossy are to play a justificatory role, they cannot be unanalyzable primitives. Instead, they must be composed of simple elements that can recur in other thoughts such as the boulder is probably damp. The boulder in both thoughts must reach out to the same boulder in the world. This kind of argument is familiar from Fodor (Reference Fodor, Stich and Warfield1994). I realize it is controversial. Even so, it is reasonable to ask of a theory of intentionality that it not rule out the possibility of explaining a propositional thought in terms of its constituent parts.

Let me sum up the dialectic so far. PI grounds intentional content in phenomenal character. We then have to wonder whether intentional content—where this is propositional and admits of a truth value—is the basic unit of aboutness. If so, then PI can help itself to an account of representing as truth or satisfaction. I argued that this is an inappropriate stopping place. Moreover, that account brings in its train a host of unattractive consequences.

The most promising version of PI will, then, have to tie subpropositional elements to their objects in the world. Note that not every such element needs to be anchored in the extramental world; syncategoremata such as ‘is’ and ‘as’ will presumably merit a very different treatment. Nevertheless, we have seen that boulder, whether it occurs in Jane's thought in the context of a proposition or on its own, is a representation in its own right.

At this stage, the representation objection becomes more pressing. In virtue of what does what-it's-like for Jane to touch the boulder represent the boulder and not the waterfall? As always, we have the option of declaring it sui generis and moving on. If there is another option, we should take it. Such an alternative would have to start by explaining representation rather than intentional content. And it would have to appeal to an internal relation. Happily or not, there is exactly one possibility left: resemblance.

2. Representation as Resemblance

Few notions have attracted as much contumely as resemblance has. Tyler Burge (Reference Burge, Pettit and McDowell1986: 128) claims that it is ‘an intuitively powerful but primitive idea’; Nelson Goodman (Reference Goodman and Goodman1972: 437) calls it ‘a pretender, an impostor, a quack’, always turning up ready to solve problems it can't (see Isaac Reference Isaac2013 for a catalog of further abuse). And yet for much of philosophy's history, it was precisely the resource philosophers used to explain representation (see De Anima in Aristotle [Reference Barnes1984: vol. 1] as well as Lagerlund [Reference Lagerlund and Zalta2011]; for the use of resemblance in historical figures, see Watson [Reference Watson1995] and Jacobson [Reference Jacobson2008] and [Reference Jacobson2013]). We have to be clear about what this ancient, if not primitive, idea is. It is sometimes said that everything resembles everything else, in some respect or other. Gloucester cheese and the square root of 75 resemble each other in that both have just been used in an example. Although quite true, this claim is true in an extremely weak sense of ‘resemblance’. The idea worth attacking (and defending) is what we might call internal resemblance, that is, resemblance in virtue of the intrinsic properties of the relata. In what follows, I shall always use the term in that restricted sense.

Although PI so far has not produced anything more than a few hints about the nature of representation, following those hints leads us to resemblance. Consider, for example, Kriegel's statement of Charles Siewert's (Reference Siewert1998) argument for the existence of phenomenal intentionality:

It is entirely possible that ‘squarish phenomenal character’ is short for ‘a state whose intentional content is true just in case there is a square in the environment’. If so, we are back where we started: talk of squarish phenomenal character does nothing to illuminate representation.

Suppose, instead, that we take ‘squarish phenomenal character’ at face value. What, then, would be the relationship between the square object and the squarish phenomenal character? A natural answer is that both, in some way, instantiate squareness. After all, what is being invoked to explain the intentionality of the phenomenal state is not its simply having some phenomenal character or other, but having precisely this character, namely, squarishness. And when two things exemplify the same property, they can be said to resemble each other in that respect. (Trope theorists are invited to substitute their own analysis of resemblance. The precise analysis of resemblance makes no difference to my account, so long as it remains an internal relation.) I am not claiming that Kriegel or Siewert would themselves accept this gloss. The possibility that they might be best served by doing so remains.Footnote ¹

Resemblance also fits neatly with Michelle Montague's (Reference Montague and Kriegel2013) ‘matching’ view. Montague argues that we achieve perceptual access to a material object when we represent enough of its properties. But how do we represent those properties in the first place? I would argue that the best answer is that we do so by instantiating them ourselves in some way. Admittedly, this view is strange, at least on first hearing, but it is familiar to historians of philosophy (it runs through Aristotle, Aquinas, and Descartes, for example), and has recently been resurrected (see, e.g., Harold Langsam Reference Langsam2011: 63f.) Obviously neither we nor our experiences can instantiate the spatial properties of material objects in the same way those objects do. Instead, the resemblance theory has to appeal to what Descartes called ‘objective being’. Much remains obscure here. But the metaphysics of this kind of instantiation are a further set of issues on which I can remain largely neutral.

There is one further possibility worth mentioning. Someone might argue that there is an isomorphism between the contents of visual experience and the physical objects that experience represents. Very roughly, the idea would be that at least some of the relations among the phenomenal ‘points’ that make up apparent shapes are preserved among the parts of the physical objects that are experienced. Such a view is to be found, in a much more sophisticated form, in the work of Chris Swoyer (Reference Swoyer1991 and Reference Swoyer1995). Despite appearances, this view should still count as appealing to an internal relation to ground representation. What flanks the representation relation will be relations among the constituents of each relatum. If an internal relation is one that obtains solely in virtue of the properties of the relata, then it does not seem to matter whether those properties are themselves relations so long as they hold within, as it were, the two objects. Hence I take the isomorphism account to be a possible option within the resemblance family, rather than a competitor.

Having bracketed the question of naturalism, I take it as no objection that the proposal fails to meet that demand. Like many defenders of PI, I am happily shirking my physicalist duties. And I am helping myself to the notion of properties being instantiated in experience. But note that doing so does not rob the proposal of its explanatory power. In fact, as long as our goal is, with PI, to explain aboutness by means of an internal relation, resemblance is the only game in town.

But how exactly is resemblance supposed to help? Recall that part of the evidence for phenomenal intentionality is the accuracy condition implicit in phenomenology. But merely undergoing an experience with a squarish character is not enough to represent an object in the world as being square. As the moderns liked to put it, simple ideas are neither true nor false.

I think this consideration shows that PI needs an additional mechanism, beyond resemblance, to explain representation. The basic idea is due to Descartes (Reference Descartes, Cottingham, Stoothoff and Murdoch1984), but has recently been brought up to date by Katalin Farkas (Reference Farkas and Kriegel2013). On Descartes's view, we think of our sensations as ‘of’ an external world only because, at an early stage of development, we gained the ability to move, to twist ‘around aimlessly in all directions in. . . random attempts to pursue the beneficial and avoid the harmful’ (Reference Descartes, Cottingham, Stoothoff and Murdoch1984, vol.1: 219). Only at this point does the child realize that the objects it pursues or avoids have an existence outside itself. Similarly, Farkas argues that experiential states on their own do not ‘point’ to anything outside themselves. In this respect, they are like pain, on some accounts. Instead, what explains the object-presenting feature of experience is the structure of such experience. As Farkas puts it, it is ‘the way these [phenomenal] features hang together and respond to movement and inquiry’ that accounts for our taking them to be of objects (Reference Farkas and Kriegel2013: 100). The phenomenal features are not piled on top of each other at random. They stand in systematic connections with each other and with the actions of the subject.

This Cartesian move seems to violate transparency. G. E. Moore, himself a sense datum theorist, might have been the first to claim explicitly that ‘when we try to introspect the sensation of blue, all we can see is the blue: the other element is as it were diaphanous’ (Reference Moore and Moore1903: 25). Nothing about such apparent transparency commits us to any particular story about representation (as Tye Reference Tye2000: 47 in effect points out). In fact, that our experiences should be transparent makes sense, if they have the kind of structure Descartes and Farkas point to.

One might also worry that I have given away the store by appealing to external relations to explain the object-presenting nature of experience. Note, first of all, that those relations hold among phenomenal states, not between such states and the world beyond the mind. What is more important, however, is that the Cartesian account does not use the structure of experience to answer the chief question we're concerned with. We've been worried about question (2): in virtue of what does this phenomenal character represent this object or property? Instead, the Cartesian account is best construed as an answer to the adjacent question (1): what makes a state a representation at all?

Let me close this section by illustrating the current proposal. Suppose Jane is standing in front of black cube on a white floor. Her visual experience instantiates the property of cube-ishness. It does so, of course, by also instantiating color. Without committing ourselves on the ontology of color, we can say that the experienced color itself has a shape, and it is this shape that resembles the shape of the cube itself. Just as PI requires, it is the phenomenal character of Jane's experience—its cube-ishness—that grounds the aboutness of the mental state that has that character.

3. A Defense of the Resemblance-based Account

Someone might well accept my arguments so far and conclude: so much the worse for PI. If the only way for PI to answer the representation objection is by invoking resemblance, then PI must be rejected. In this section, I argue that this conclusion is too hasty. The typical objections philosophers lodge against resemblance theories are not nearly as powerful as they seem. What is more, resemblance cum phenomenal intentionality provides resources for answering those objections that resemblance on its own lacks.

1. (i) Ubiquity. For any a, there will be a large—though not infinite—number of intrinsic properties it possesses, and, plausibly, there will be at least one b such that b has at least one of those properties. Resemblance is all over the place; representation isn't. So they cannot be the same thing. (See, e.g., Burge Reference Burge, Pettit and McDowell1986: 128 and Meyering Reference Meyering1989: 3)

The best reply is a concession. One car resembles another; it would be absurd to say that each therefore represents the other. Representation cannot just be resemblance. But we have to keep in mind which question we're trying to answer. Throughout we've had to distinguish the question (1) what makes a state a representation from (2) what makes it represent this rather than that? To be taken seriously, the resemblance theory has to be an answer to (2) rather than to (1). A state has to do more than resemble something to become a representation. In answering (1), the resemblance view has two resources. First, it can hold that only states with phenomenal character are representations. Although not everyone aligned with PI will accept this, some will. More plausibly, the resemblance view can point to the second step in achieving representation: the subject's experiences have to exhibit a certain kind of structure before they become representations. Either of these moves (and there are surely more) would make representation much more scarce than resemblance.

But there is a stronger version of the ubiquity problem lurking: what Satosi Watanabe (Reference Watanabe1969: 376) calls the ‘Ugly Duckling Theorem’. Suppose that the degree of similarity between a and b is fixed by the number of predicates that can be applied to them. Watanabe argues that any object will bear precisely the same degree of similarity to any other object. The ugly duckling is just as similar to the fetching cygnet as is any other cygnet. Applied to the present theory, the claim would be that any given mental state will resemble any object to the same degree as any other object. Despite appearances, the experience with the cube-ish phenomenal character does not resemble the cube any more than it does the sphere.

Before meeting the objection directly, let me offer some reasons for being suspicious of the theorem, at least as applied to the present view. It is significant that Watanabe's argument for the theorem appeals to predicates, not properties. And it appeals to predicates that few would agree map genuine properties. Consider Watanabe's (Reference Watanabe1985: 75) version of the argument: ‘To persuade that two objects are similar, it is natural to enumerate the properties that are commonly owned by the objects. . .. For such an argument it is only fair for the arguers to be allowed to use all possible predicates that are applicable to the objects’. Note that Watanabe begins by talking about properties but shifts to predicates. Allowing any applicable predicate at all to count toward similarity makes proving the theorem all too easy, as does permitting external points of resemblance (say, ‘being north of the Equator’). But the view I am defending appeals to internal resemblance in respect of properties, taken as real members of the mind-independent world, not in respect of mere predicates. Hence it is not obvious that Watanabe's theorem threatens the resemblance view, properly understood.

But let us waive these considerations. What is needed to answer the objection directly is some way of weighting the properties in respect of which the similarity obtains. And here is precisely where the resemblance account's marriage to PI pays dividends. The salient features are precisely those intrinsic properties that capture our attention. One notices the shared shapes of the comely cygnets and notes their difference from that of the duckling. There are no doubt plenty of similarities that we simply do not attend to even if they somehow feature in our phenomenal experience. In fact, something very like this reply is common currency in cognitive science. As Stevan Harnad puts it, in his reply to Watanabe's argument: ‘But of course our sensorimotor systems do not give equal weight to all features: they do not even detect all features. And among the features they do detect, some (such as shape and color) are more salient than others (such as spatial position and number of feathers)’ (Reference Harnad, Cohen and Lefebvre2005: 32). Replace ‘sensorimotor’ with ‘phenomenally perceptual’, and you have a fair approximation of this point. In short, the Ugly Duckling theorem tells against the present account only if it assumes that there is no way to privilege some of the features shared by experiences and objects over others. And that assumption is false.

1. (ii) Symmetry. Representation is asymmetrical: if a represents b, then, at least typically, b doesn't represent a. Resemblance, by contrast, is symmetrical: each Winklevoss twin resembles the other.

Here again the difference between questions (1) and (2) makes all the difference. True, resemblance is symmetrical and representation is not. So while my experience of a cube resembles the cube and vice versa, it would be absurd to say that the cube represents my experience. Why it is absurd depends on just how we answer question (1). Suppose that only mental states can be representations. Then the reply is clear: the representation relation is asymmetrical not because resemblance goes in only one direction—it doesn't—but because the cube is not a representation at all.

1. (iii) Singularity. Suppose I am looking at a cube; my experience and the cube share a property, cube-ishness. But cube-ishness is shared by every cube in the actual world and out of it. How does my experience now get tied to this cube and not all those others, actual and possible?

It's important not to exaggerate the threat posed by objection (iii). If we limit ourselves to features of experience that are accessible from the first-person perspective, then any view will have to tolerate some degree of unspecificity. And one of PI's core commitments is the subjectivity of content. Such specificity as there is must be tied to introspectible features such as the distribution of color, size, shape, and position. In principle, a single experience can always be an experience of more than just the particular environment in which it happens to occur. If this is a problem, it is one any adherent of PI will have to live with.

But remember that the resemblance account under consideration can use the resources of PI. It is true that, taken singly, a phenomenal experience that instantiates, say, black-cube-ishness will likely be indistinguishable from another such experience. Fortunately, we hardly ever take our experiences singly. Jane's phenomenal world nearly always includes more than a single phenomenal property, such as the feel of the boulder; she also experiences the color of the moss and the variegated shading of the trees. It is not as if it is up to just one of these aspects of her experience to tie that experience to the world. Instead, all of them can work together, at least to the extent that they are objects of her attention. If we keep this fact in mind, the worry that Jane's experience is insufficiently singular should at least be greatly reduced or even vanish. For the number of subjectively identical ‘experiential scenes’ is far fewer than the number of subjectively identical squareness-es. The final step is achieved by looking not just at a single slice in time, but at Jane's experience over a period of seconds or even minutes. For part of what knits Jane's experiences over time together with her environment is surely the fact that the properties instantiated in the objects around her, both synchronically and diachronically, are also instantiated in her experiences.

Finally, note that objection (iii) should not be mistaken for the claim that singular thoughts are such that they depend for their existence on the existence of their objects (see McDowell Reference McDowell1998). Whether or not there are such object-dependent thoughts, anyone committed to the narrow intentionality of phenomenal states will deny that perceptual states are among them.

1. (iv) Any story about representation has to account for misrepresentation. But if a resembles b, a cannot misrepresent b on pain of not being a representation of b at all.

Suppose we take resemblance to be (literal) property-sharing. If a resembles b in virtue of the fact that Fa and Fb, a represents b just as b in fact is, at least in respect F. It seems that the only way for a to misrepresent b is for it to lack F, but since representation is resemblance, then it is no longer a representation of b.

The present view has a ready response. Suppose the felt shape of Jane's experience fails to resemble the shape of the boulder underneath her hand, due to some sort of haptic illusion. There will still be other qualities of the boulder that her experience gets right, such as its distance from her, its contact with her hand, and so on. The degrees of misrepresentation correspond to the number of respects in which Jane's experience resembles the world.

But now suppose Jane's experience is as of a pink elephant (to borrow Montague's example [Reference Montague and Kriegel2013: 39]). Now none of the qualities instantiated in her experience resembles the qualities of the boulder. In such a case, there's no object in the world that she's misrepresenting. That in fact sounds like the right result, not a counterexample.

One thing any story about experience wants to preserve is this: when Jane hallucinates, she can undergo an experience that is the same, from her perspective, as one she would undergo were the world as she represents it. The present view is entitled to this counterfactual, which is made true by the intrinsic nature of her experience. Now, Jane might hallucinate in a way that just happens to match the way the world is. On the view we're considering, she is not misrepresenting the world in that case; nevertheless, we might want to deny that her experience is veridical since it is not caused by the world itself. Not all epistemic failures are failures to represent the world as it is.

1. (v) If resemblance is property-sharing, then typical visual experiences will not resemble, and hence not represent, material objects. As the visual arts demonstrate, the apparent spatial properties of objects very rarely coincide with their real ones. The properties instantiated in Jane's experience are viewer-dependent properties that cannot be identified with the viewer-independent properties of real objects.

Imagine Jane watching a hawk flying overhead. As it comes closer, it takes up more and more of Jane's visual field. And yet Jane's experience typically doesn't represent the hawk as growing in size. The ‘typically’ is important here, for with attention we can notice the shift in experienced size.

To use Gibson's (Reference Gibson1950) terms, my claim is that Jane can experience both a visual world and a visual field. It's very unlikely that she experiences both at the same time, all the time. In fact, as with figure/ground illusions, it might be impossible for her to instantiate both sets of properties at once. In typical conditions, Jane's experience will amount to experience of a visual world. Only in special circumstances (such as when reflecting on optical illusions or trying to draw something) will Jane experience her own visual field.

How does any of this help with our present problem, namely, the seeming mismatch between viewer-dependent and viewer-independent properties? The objection assumes that the only features instantiated in experience are dependent on the viewer. What we've just seen, however, is that there is another layer to experience. There are also features instantiated in experience that do resemble the mind-independent world. What Jane attends to, in the usual case, is the visual world, which co-instantiates many of the properties of the actual world.

I have no doubt other objections will spring to the reader's mind. Answering all of them would require fully developing the resemblance account, which I do not have space to do here. And like other participants in PI, I have sailed blithely past the whole issue of naturalism. My goal, however, is to open a spot, however small, for the resemblance account in the logical space of the debate. Readers who think there is no such spot available are invited to regard the argument of the paper as a reductio of PI. For whether it succeeds or fails, the resemblance account is the best answer PI has to the representation objection.