Jump to ContentJump to Main Navigation
LOT 2The Language of Thought Revisited$

Jerry A. Fodor

Print publication date: 2008

Print ISBN-13: 9780199548774

Published to Oxford Scholarship Online: February 2010

DOI: 10.1093/acprof:oso/9780199548774.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2017. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in OSO for personal use (for details see http://www.oxfordscholarship.com/page/privacy-policy). Subscriber: null; date: 26 February 2017

The Metaphysics of Reference

The Metaphysics of Reference

(p.196) 7 The Metaphysics of Reference

Jerry A. Fodor (Contributor Webpage)

Oxford University Press

Abstract and Keywords

Only matter has causal powers. The underlying metaphysical intuition is perhaps clear enough to be getting on with: whatever enters into causal interactions is constituted of the sort of stuff that basic physics is about. Call that the ‘physicalist’ thesis (PT). This chapter argues that PT functions as an a priori methodological constraint on scientific practice; ‘a priori’ in the sense that any theory that fails to conform to PT to that extent counts as disconfirmed. This applies to intentional psychology inter alia: only matter can think. It also discusses what philosophers call the ‘naturalization’ of intentional psychology.

Keywords:   metaphysical intuition, physicalist thesis, naturalization, intentional psychology

Only matter has causal powers. I don't know exactly what that means, but the underlying metaphysical intuition is perhaps clear enough to be getting on with: Whatever enters into causal interactions is constituted of the sort of stuff that basic physics is about. Call that the ‘physicalist’ thesis (PT); I suppose it's at least part of what one has in mind when one speaks of basic physics as basic.1 The development of science has been driven by the physicalist thesis at least since Lucretius. Unlike many of my colleagues, I think PT functions as an a priori methodological constraint on scientific practice; ‘a priori’ in the sense that any theory that fails to conform to PT to that extent counts as disconfirmed. This applies to intentional psychology inter alia: Only matter can think. If I believed in implicit definitions and the like, I would say that the physicalist thesis partly defines the scientific enterprise as we have come to understand it so far. But I don't, so I won't.2

Now, the main goal of this book is to help make science out of commonsense intentional psychology (my life's work, I (p.197) suppose, if you don't include feeding the cat). So, I'd better say something about what philosophers call the ‘naturalization’ of intentional psychology. (For present purposes, a naturalized psychology, intentional or otherwise, is just one that comports with PT.)

What makes intentional psychology problematic from the point of view of the physicalist thesis is, of course, its commitment to intentional states and processes; that is, to states and processes that are both endowed with causal powers and susceptible of semantic evaluation in respect of their truth, reference, aboutness, content, and the like. It is not immediately obvious that something with those sorts of properties could be constituted entirely by the sorts of things that the ontology of basic physics recognizes.3

But, despite its having a plausible claim to centrality, PT actually isn't much discussed these days either in the philosophical literature or in cognitive science. Not in cognitive science because, in their deepest heart of hearts, the great majority of psychologists think that talk of truth is unscientific (not to say vulgar) and that they already know how to naturalize reference. Namely, it's got to turn out to be some sort of association, either among ideas or between ideas and the world (or between stimuli and responses in case the psychologist in question is a behaviorist); and it's not hard to imagine a machine that can form associations. Imagining such machines is what connectionists do for a living.

By way of ironic contrast, philosophers, at least since the Wittgenstein disaster, don't worry much about naturalizing intentional psychology because they think it's known to be impossible. (This is one of the very important ways in which the mainstream of (p.198) contemporary ‘analytic’ philosophy departs from the historical mainstream of empiricist theorizing.) What makes the naturalization of intentional psychology impossible is, putting it very roughly, that semantic content inheres, in the first instance, in the expressions of public languages,4 and their metaphysics is essentially connected to conventions, norms, and the like. Since conventions, norms, and the like are themselves up to their ears in intentionality, the application of PT to intentional explanations couldn't but breed circularities. The suggested moral is either that intentional explanations constitute an ‘autonomous’ realm of discourse or that they are merely façons de parler eventually to be displaced by discourse that is explicitly about brains. In neither case could intentional psychology be a science in the sense that geology is.5

The unsurprising consequence of these ways of thinking is that, even by the usual standards, the geography of this part of the woods is very badly understood. It is even depressingly unclear just what problems a successful naturalization of intentional/semantic properties would have to resolve. That being so, I see no way to proceed except by assuming lots of things, and what I'll have to say about the naturalization of the intentional/semantic is unlikely to be of much interest unless the assumptions are correct. Oh well.

Some assumptions.

  1. 1. To begin with, I assume, in the spirit of the preceding chapters, that the semantics of thought is prior to the semantics of language. So, for example, what an English sentence means is determined, pretty much exhaustively, by the content of the thought it is used to express. The corresponding implication is that semantics is essentially a theory of thoughts, the contents of which are, I suppose, not determined by norms and conventions. Quite possibly (p.199) English has no semantics, some appearances to the contrary notwithstanding.

  2. 2. I assume that reference is compositional; the reference of a complex expression is a construction out of the referents of its parts. (In fact, something stronger; reference is reverse compositional: see Fodor and Lepore 2002).

  3. 3. I assume, in the spirit of the preceding chapters, that referentialism is true. There is no problem about naturalizing meanings, senses, and the like because there aren't any such things.6

  4. 4. I assume that there are two kinds of reference: reference to individuals and to properties. This means, from the syntactic point of view, that the vehicles of reference are exhaustively singular terms and predicates.

  5. 5. I assume that some sort of causal theory of reference is true, both for predicates and for singular terms.7 The residual question is then: ‘which sort?’. The rest of this chapter is a claim that some of the most important standard arguments purporting to show that there can't be a causal theory of reference are unsound. (If I'm right about that, then perhaps one should stop assuming in the philosophy of language and/or the philosophy of mind that naturalization isn't possible.)

  6. 6. I assume that the crux of the problem of naturalizing reference is to provide a theory of perceptual representations. The (p.200) paradigms of such might well be present‐tense demonstrative thoughts (thinks: That's a cat). Having got a substantial number of cases of perceptual reference under control, the rest of the story might appeal to some or other sort of definite description to fix the reference of Mentalese terms that don't occur in present‐tense perceptual thoughts (the cause of that grinding sound; the guy I saw in the kitchen yesterday; and so forth).

SNARK. ‘Cheater, cheater | Pumpkin eater.’ AUTHOR. Really, Snark; at your age. And with so respectable a publisher. SNARK. Sorry; it slipped out. But I had thought you said there are at most hardly any definitions. AUTHOR. I had thought I said that too. SNARK. Then tell me, please, how you get from a vocabulary that is (by assumption) only rich enough to refer to (some? many? whatever?) things that are current objects of perception to a vocabulary that is sufficient for reference to things that are not current objects of perception? Tell me that. Go on. I can hardly wait. AUTHOR. Suppose E is an expression in a vocabulary of the latter kind. You're quite right that I can't assume that it has a definition in a vocabulary of the former kind. I can't, given the things I've been saying, assume that it has any definition at all. But I don't need to if the metaphysics of reference is to be understood in terms of the causal locking of a term to its referent. All I need is something in the perceptual vocabulary that I can use to establish the locking. It might, for example, be a definite description of the referent of E that is (as one says) ‘rigid’.8 If so, then by definition it isn't a definition of E.

(p.201) In fact, come to think of it, what locks E to its referent needn't even be a description that's true of E's referent. It's pretty widely recognized that new expressions can be introduced by descriptions that are false of their referents (see especially Donnellan 1972). My guess is that that sort of thing happens all the time. In particular, my guess is that that sort of thing is at the heart of our capacity for transtheoretical reference. It explains why I can use the vocabulary of my theory to criticize assertions that you make in the vocabulary of your theory even if (as is surely the usual case) neither of our theories is actually true. I don't imagine you could do science lacking some such arrangement. Or anything else much either. This is as far as you can get (at least, it's as far as I can see how to get) from the view that the semantics of a theoretical term has to be relativized to the theory that it's a term in (either in the manner of Kuhn or in the manner of Ramsey). Good. Semantic relativism, besides being false, is a pernicious and immoral doctrine which I would not wish even on a snark.

Well, so I'm making these six assumptions,9 all of which (except perhaps the last, which is loosely in the tradition of empiricism10) are tendentious. Some of them are very tendentious. So be it. With all of them in force, we turn to the main business of this chapter, which is to consider the prospects for constructing a naturalized, causal theory of reference. To my knowledge, there are three prima facie objections to the project which, severally and collectively, have convinced many—maybe (p.202) most—philosophers that it's a dead end; there may be others, but these are iconic.

The first I've already mentioned; namely, the putative groundedness of content, including referential content, in a metaphysics of norms, conventions, and the like. It suggests that a causal account of reference would presuppose the naturalization of intentionality at large, which, though perhaps not impossible in principle, is also not likely to be achieved by next Tuesday. In any case, I'm interested in the idea that a naturalistic metaphysics of content proceeds from the reduction of reference to causation to a general, naturalistic treatment of the intentional/semantic; if so, the former mustn't presuppose the latter.

The second familiar worry involves not naturalistic theories of reference in general but causal theories of reference in particular. It concerns a tangle of issues about how, if reference is a kind of causation, it's possible to think about, or speak about, anything except the current cause of one's thought or utterance. The paradigm here is what I'll call the problem of ‘wild tokenings’ (of which the notorious ‘disjunction problem’ is a special case).

Finally, there is the ‘Which link’ problem: If referents are in some way causes of the utterances/thoughts that refer to them, then each of the former must belong to a chain of events that eventuates in the tokening of one of the latter. But causal chains extend (presumably) indefinitely in both directions. So, the question arises what determines which link in such a chain of events is the referent of the thought or utterance.

I wouldn't bother you with these worries except that the project of this book is, after all, to help make respectable science out of content, and I do think (see Introduction) that naturalism is part of what the respectability of a science consists in. That being so, I would feel better and sleep more soundly if there are prima facie replies to the prima facie arguments against the reduction of reference to causation.

(p.203) 7.1 Normativity

Everybody goes on about norms. Cows go ‘moo!’, philosophers go ‘norm!’. The idea, roughly, is that contents of symbols emerge from conventions that control their use, thereby determining what it is to use them correctly. Since causation per se is neither correct nor incorrect, content can't reduce to causation; and that applies to referential content inter alia.

But though this line of thought is simply ubiquitous in the literature of analytic philosophy (cf. Wittgenstein, Sellars, Davidson, Dennett, Brandom, etc.), we can safely ignore it. For better or worse, we're committed to LOT; and LOT, though it is a system of representations, isn't a system of representations that anybody uses, correctly or otherwise. One doesn't use thoughts, one just has them. Having thoughts isn't something that you do; it's something that happens to you. There are speech acts (maybe); but there aren't any thought acts. And so forth. These are all ways of saying much the same thing; they're part and parcel of the idea that LOT is a system of representation that mediates thought, not of one that mediates communication.

SNARK. Doesn't that make LOT a private language? And didn't Wittgenstein prove that there can't be such a thing? AUTHOR. I think that what Wittgenstein meant by a private language is one the referents of whose expressions are epistemically accessible only to its user; and, of course, LOT isn't one of those; we use the concept COW to think about cows; and cows are public objects. If, however, Wittgenstein did use ‘private language’ to mean something like LOT, then I can't imagine what he had against them. They look all right to me.

(p.204) 7.2 ‘Wild Tokens and the Disjunction Problem’

Purblind on a dark night, John mistakes a large cat for a small cow. He thinks, as it might be, there goes a small cow. So his having the thought is, inter alia, an instance of his thinking of something as a cow; in particular, it is not an instance of his thinking of something as a cat or as a cat‐or‐cow).11 Or, in the course of a daydream about cows John comes to have a thought about cats; so a thought about cats is caused by a thought about cows, but neither cats nor cows are causes of either. How is this possible if reference reduces to causation?

I haven't much to say about the ‘wild tokens’ problem that I haven't said elsewhere. Here's the basic proposal (for lots of discussion see Fodor 1990; Loewer and Rey 1991): By assumption, if John is referring to anything at all, he is referring to something that caused his thought; and, of course, everything caused by a cat is ipso facto caused by a cow‐or‐cat. Thus the disjunction problem. But there is (so I claim) nonetheless a difference between the case where John is thinking of a cat as a cow‐or‐cat and the case where John is thinking of a cat as a cow. It resides in the counterfactuals: John couldn't have thought of a cat as a cow‐or‐cat unless he could have thought of it as a cat and could have thought of it as a cow. But this doesn't go the other way around: being able to think of a cow as a cow or as a cat does not require being able to think of it as a cow‐or‐cat. So, being able to think about cows‐or‐cats is, as I've sometimes put it, ‘asymmetrically dependent’ on being able to think about cows and being able to think about cats. Because of this difference in the counterfactuals (so I further claim), a causal theory of content can, after all, accommodate our intuitions (p.205) about the content of John's thought: Since the concept that John (mistakenly) applied to a cat was one that he could have had even if he didn't have the concept CAT, it must have been a token of the concept COW that John's encounter with the cat caused him to entertain, and not the concept COW‐OR‐CAT. So everything comes out right. (Unless I've made a mistake. Which is perfectly possible.)

Suppose this sort of treatment really does capture our intuitions about John. The question remains whether it works for wild tokens at large. I have my suspicions, but I'm not going to foist them on you. Further research is, as one says, required. Still, if I've got it right about John, that would seem to put a dent in what has been widely considered an impregnable argument against such theories; namely, that they can't (indeed, demonstrably can't) provide an account of what it is to say or think something false (to misapply a concept, etc.). It is, I'm suggesting, simply a fallacy to argue that since there can't be such thing as a miscause, there likewise can't be such a thing as a causal theory of mistakes. And, notice that although ‘mistaken’, ‘incorrect’, and maybe even ‘false’ are indeed normative notions, the sort of treatment of the disjunction problem that I've suggested doesn't invoke the norms in question. Or any others.

7.3 Which Link in the Causal Chain?

It's common ground that any instance of a perceptual thought comes at one end of a causal chain. If such a thought refers at all, then, if the causal theory of reference is true, we must suppose that what the thought refers to is some or other link in such a chain of causes. The problem is: Which link?

The sequence of events that causally connect ones thoughts about cats to cats typically includes, for example, one's cortex becoming active in various interesting ways. Now, one might be prepared at a pinch to live with a little semantic indeterminacy if that's the price of a naturalized theory of reference. But ‘There's a (p.206) cat’ refers to a cat, not to a cortex; if that's not common ground, what would be? Worse still: For all I know, the causal chain that leads to one of my CAT tokens reaches all the way back to the big bang; for all I know, every causal chain there ever was reaches all the way back to the big bang. If so, it may seem that no causal theory can rule it out that the big bang is all that anybody ever thinks about. But, surely, nobody thinks about the big bang all of the time; not even Woody Allen. Something must be done.

Here's where we start to get some use out of the assumption that the core cases we have to deal with are perceptual thoughts (more or less equivalently, that they are demonstrative thoughts, since, I suppose, what you're in a position to perceive is pretty much what you're in a position to demonstrate). So, the big bang is out as the referent of my thoughts about Mr James the cat; and so is everything that happened yesterday and so is everything that will happen tomorrow; and so is everything that didn't happen; and so forth. None of these is currently perceptually available. Also, as we'll see presently, if the core cases are perceptual, that willy‐nilly drags in the notion of a perspective, and that helps too. What you can (currently, visually) perceive depends on what's visible from your current point of view. Things beyond your visual horizon are out, and so are things that are within your horizon but that are eclipsed by things between you and them.

So, for various reasons, we can picture you as being situated at the center of a circle that includes all but only the things you can now see from here, and as being at the end of a causal chain which intersects the circumference of that circle. By assumption, whatever your current perceptual thought refers to must be among the links of that chain that are in the circle. The question, to repeat, is: Which such link? The best we've done so far is to reduce the number of candidates.

Actually, I think a notion that Donald Davidson introduced (for, however, a rather different purpose) provides the answer. That I do think this rather surprises me, since I haven't, by and (p.207) large, much sympathy with Davidson's views about the content of beliefs, perceptual or otherwise.

7.4 Triangulation

As I read Davidson, he was deeply involved in trying to find a transcendental argument that there can be no intentional content without (at least the possibility of) radical interpretation;12 a thesis from which, as I remarked above, the autonomy (irreducibility, whatever) of the intentional follows pretty directly, as does the claim that languages must be epistemically public objects.

I think Davidson thought he might ground the argument he wanted in the epistemological conditions for radical interpretation: Any content that an informant's utterance(/thought) can have must ipso facto be accessible to a ‘radical interpreter’; that is, to an interpreter who starts off with nothing but (say) a priori principles of logic and methodology, and whose empirical data consist of information about the informant's observable interactions with the environment in which he is situated.13 I think Davidson thought he could rely on this because interpretation from the radical interpreter's epistemic position would be a condition for a language to be learnable, and he took for granted that learning its semantics is an essential part of learning one's first language. In short, since the epistemic position that the radical interpreter is supposed to occupy is presumed identical to the epistemic position from which children actually do learn language, it must be possible to learn a language from that position. All that being so, any transcendental consequence of the possibility of radical interpretation must be (p.208) true. (I once asked Davidson why only semantics, among all the various empirical disciplines, labored under this sort of a priori epistemic constraint; it isn't plausible, after all, that whether there's a cave in the middle of a mountain depends on whether someone in the epistemic position of a radical geologist could find one there. Davidson replied that ‘language is special’. I suppose what he meant was that theories about languages are tied to theories of learnability in a way that geological theories aren't. Taking it for granted that there is some epistemologically interesting sense in which first languages are learned has brought many a philosopher and many, many a psychologist, to grief. See Ch. 5.)

Let's just stipulate that all that is more or less correct, or that it could be made more or less correct by appropriate tinkering.14 Now, since the sort of theory of radical interpretation that Davidson had in mind was a species of causal theory of reference, it has the usual ‘Which link?’ problem (as, indeed, Davidson fully recognized). How, then, is that problem to be solved on an account of content that says that the possibility of content entails the possibility of radical interpretation?

Suppose I am a linguist bent on interpreting Adam's utterances on a certain occasion when we are both puttering about in the garden. Suppose that a large, hairy, ugly snake suddenly swings into view.15 (To simplify, we may stipulate that nothing else of interest happens at the same time.) Adam thereupon emits an utterance in his native tongue and I likewise emit an utterance in mine. The scene is now ripe for what Davidson called ‘triangulation’.

Here, very approximately, is how I take it that triangulation is supposed to work. I'm aware that the vocable I uttered in the situation imagined means (in my home language) something like Go away large, hairy, ugly, snake. Indeed, the intention with which (p.209) I uttered that form of words was precisely to express my desire that the large, hairy, ugly, snake in question should go away.16 So, what I was talking about lies on a causal chain from the snake to me, just as a causal account of reference requires. OK so far. But notice that the same sort of reasoning that I've just applied to me qua interpreter also applies to Adam qua informant; or at least it's reasonable for me to suppose that it does in my attempt to interpret Adam's utterance. After all, given that Adam is much the same sort of creature I am, it's thoroughly plausible that, like me, he said what he did with the intention of expressing the desire that the snake in question should go away. All that being so, it's thoroughly reasonable for me to suppose that the form of words Adam uttered is one that speakers of his language use when a desire to be rid of a snake comes upon them.

So I can triangulate. That is, I can reasonably suppose that a causal chain that runs from the perceptual horizon to my utterance would intersect a causal chain that runs from the perceptual horizon to Adam's utterance, and that it would do so at the snake. Put very schematically, the idea is that the link that's the object of a speaker's thought(/the referent of his utterance) is the one where the causal chain from the world to the speaker(/thinker) intersects the causal chain from the world to his interpreter when the speaker and the interpreter are caused to utter tokens of the same type. Ceteris paribus, to be sure.

The essence is that, in the situation imagined, I have a lot more information available with which to interpret Adam's utterance than had at first appeared. Together with my observation of the form of words he used and the circumstances in which he used it, I also have what I know about my own reaction to being in the situation that he was in. Because I do, I can infer what Adam is likely to have said by using the form of words that he did, and hence what that form of words is likely to mean in his (p.210) language:17 It's likely to mean the same thing that my utterance in my language did; namely, Go away large, hairy, ugly snake. And, given that Adam and I are both referring to the snake, the causal theory allows me to infer that the snake is on both the causal chain that leads to his utterance and the causal chain that leads to mine. In particular, it's at the link at which those chains intersect. Were this chain of inference unsound, there would be (in point of both metaphysics and epistemology) no way that I could have interpreted what Adam said, and no way that a child in my position could have figured out what the form of words Adam uttered means; quite generally, there could be no matter of fact about the referential semantics of Adam's language. So I suppose Davidson to suppose.18 ,19

Here are some characteristic passages.20

This much should be clear [sic] the basic triangle of two people and a common world is one of which we must be aware if we have any thoughts at all. If I can think [at all] I know that there are others with minds like my own, and that we inhabit a public time and space filled with objects and events . . . (p. 86) The triangle I have described stands for the simplest interpersonal situation. In it two (or more) creatures each correlate their own reactions to external phenomena with the reactions (p.211) of the other . . . interaction, triangulation . . . gives us the only account of how experience gives a specific content to our thoughts. Without other creatures with whom to share responses to a mutual environment, there is no answer to the question what it is in the world to which we are responding . . . (p. 129) The child finds tables similar; we find tables similar. It now makes sense for us to call the responses of the child responses to tables. Given these patterns of response, we can assign a location to the stimuli that elicit the child's response . . . It is a form of triangulation: one line goes from the child in the direction of the table, one line goes from us in the direction of the table, and the third line goes between us and the child. (p. 119)

There's much more of the same kind in the Davidson corpus, but that should suffice to give the feel of the thing.

Well, as the reader will have predicted, I think just about every word of that is false. In particular, I don't think that learning a language has anything much to do with correlating linguistic responses to publicly available stimulations; and I doubt that there is, or could be, a theory of interpretation in anything like the sense that Davidson appears to have in mind. And (this is important) I don't see why any actual interpreter has to be involved in the process of fixing meanings. Wouldn't a counterfactual interpreter who shares Adam's psychology do just as well?21 In fact, wouldn't it do for this other person to be Adam himself, only situated in counterfactual circumstances?22 In which case, the a priori demand for an interpreter, even if it can somehow be justified, wouldn't suffice to show that the ontology of interpretation must require more than one‐person states. In particular, referential content might be no more intrinsically ‘public’ than is talking to oneself.

(p.212) We'll return to this last line of thought presently. Suffice it that, although what Davidson says about interpretation depending on triangulation appears to leak in several places, I think that something like triangulation may help solve the ‘Which link’ problem; which is, as we've seen, a problem that any viable causal theory of reference, Davidsonian or otherwise, must contrive to solve somehow or other.

OK. So there's Adam and there's the snake, and Adam utters, as it might be ‘Gavasnake’. I'm taking for granted (as per assumptions previously announced) that Adam's utterance expresses a perceptual belief and that the perceptual belief it expresses refers to a link in a causal chain that starts at Adam's perceptual horizon and eventuates in his utterance. To which such link does it refer? Suppose it refers to the snake, or to the snake's appearance on the scene.23 What, metaphysically speaking, determines that it does? I do no more than follow Davidson (though not, I imagine, to any place where he would wish to lead).

Suppose that we represent the causal history of Adam's utterance by a line that runs to it passing through its referent intersecting the perceptual horizon somewhere or other. Call this ‘Adam's line’. Every point on Adam's line corresponds to a (more or less remote) cause of his utterance. The present question is: which of the links in this causal chain is its referent? I take Davidson's intuition to be that the answer depends on the value of at least one further parameter; a parameter that isn't (fully) determined by Adam's line. And, notwithstanding that I think much of what Davidson says about triangulation isn't true, I think this intuition is spot on. To (metaphysically) determine the referent of Adam's thought, we need another line.

(p.213) Which other line? Well, to start by tidying up a bit; since we are assuming an RTM model of perceptual belief, we can stop worrying about the referent of what Adam said and ask about the referent of the mental representation a tokening of which (in his ‘perceptual belief box’ as it were) caused Adam to say it:24 Call this ‘Adam's representation’. And, since we are trying to fix the referent of a perceptual belief, we can speak literally of Adam's perspective on the scene that confronts him. Call that ‘Adam's perspective’.

Here, then, is the proposal in a nutshell: Imagine there is not just the actual Adam with the perspective that he actually has, but also a counterfactual Adam (‘Adam2’) who is, say, three feet to the actual Adam's right. Adam2 has a (counterfactual) perspective on the (actual) visual scene; one that differs from Adam's perspective in accordance with the usual (i.e. the actual) laws of parallax. Assume that Adam2 tokens a representation of the same type that Adam does. Draw a line that starts at Adam2’s token and represents its causal history (i.e. the causal history that Adam's token would have had if Adam had been at the position that Adam2 occupies in the counterfactual scenario). Call this Adam2’s line. The metaphysical problem is: given the two causal histories, solve for the referents of the tokens. RTM allows us to do so. It says that the two tokens have the same referent iff Adam's line and Adam2’s line intersect at a link; and that their referent is the link at which they intersect.

So, what Adam's utterance referred to depends on counterfactuals about the causal chains that would have caused it at the various positions that Adam might have occupied when he uttered it. At a minimum, this story must work at least as well as Davidson's; if Davidson is right to claim that the (actual) Adam's chain intersects his (actual) interpreter's line at the snake, then I am justified, and for the same reasons, in claiming that Adam's actual line would likewise (p.214)

                      The Metaphysics of Reference

Figure 7.1 Adam and his counterpart triangulate on a snake. Given the causal chain that the referent belongs to, one can figure out which link on that chain the referent is. Thus, by assumption, the referent of Adam's thought is somewhere on Adam's line, and the referent of Adam 2’s thought is somewhere on Adam 2’s line. The snake is the unique link which satisfies both conditions, so it is the referent of both thoughts.

have intersected the (counterfactual) line of Adam2 at the snake. In fact, my story is just Davidson's story about triangulation, only stripped of extraneous posits. I need no interpreter as a condition of Adam's thought having content; counterfactuals about Adam himself will do. Correspondingly, I don't need to admit that interpretable languages are ipso facto epistemologically public; all that's required is that what a speaker(/a thought) refers to needn't change with (actual or counterfactual) changes in perspective. My proposal, unlike Davidson's, is thus compatible with the view that people think in a (de facto) private language. As, indeed, I fully believe that they do.

There are various caveats with which I more or less won't bother you. For example, the procedure just outlined doesn't generalize to fix the referent of beliefs that aren't perceptual. Roughly, that's because in Adam's case, but not in the general case, we can use shifts in perspective to pick out the relevant counterfactuals; the (p.215) counterfactuals that count are the ones which would be true if Adam had had a different point of view. But, as previously remarked, I'm quite prepared to assume that if the triangulation procedure can provide a rich enough vocabulary of perceptual representations, Adam's other concepts can be introduced by definite descriptions.25

I've given no argument that the ‘Which link?’ problem is the only objection to the possibility of a causal‐chain account of the metaphysics of reference that would be left over if the normativity problem and the disjunction problem were solved. Never mind; we'll deal with the others as they come along.

So where, if anywhere, does this discussion leave us? Come to think of it, where, if anywhere, does this whole book leave us? It might be useful to enumerate some of the theses that are probably false if referentialism is true and a causal account of Mentalese reference can, in fact, be given. They include just about all of the views of mind and language that are currently in fashion in either philosophy or cognitive science.

  • 1. Practically all recent philosophers to the contrary notwithstanding (in particular, all pragmatists to the contrary notwithstanding; hence, Quine, Davidson, Wittgenstein, Dummet, Sellars, etc. and their followers to the contrary notwithstanding), reference is ontologically prior to truth; which is, of course, just what the proper treatment of compositionality, productivity, systematicity, and the like independently require. Philosophers who deny that reference is ontologically prior to truth have confused their metaphysics with their epistemology. It may be that, in the course of language learning, radical interpretation, radical translation, etc. (p.216) truth‐conditions on (some, simple) sentences are established prior to establishing what refers to what. (Or maybe not. If, like me, you doubt that languages are learned, and you doubt that there could be a theory of radical interpretation/translation, you will not wish to spend much time debating this.) But nothing about ontological priorities would follow.

  • 2. There is no particular reason to believe that reference is a social phenomenon. At no point in the preceding sketch was a ‘second person’ interpreter invoked; whenever it began to look like we might need one, it turned out that the same speaker(/thinker) in a counterfactual situation would do just as well. It is therefore unsurprising that some languages turn out to be private; what's perhaps surprising is that not all of them do.26

  • 3. There is no particular reason why supposing that mental representations have semantic properties should lead to a homuncular regress. Excepting perhaps neurologists, nobody ever interprets mental representations. Nor, in any case, could such interpretations be what bestows content on mental representations. The content of a mental representation is its referent, and what fixes its referent is the character of its causal connections to the world.

  • 4. There is no particular reason to believe that the content of expressions in public languages is metaphysically prior to the content of propositional attitudes. In fact, there's every reason to think that it mustn't be, on pain of making an utter mystery of how public languages could be learned. Generally speaking, learning requires thinking. Wittgenstein's (and Skinner's) suggestion that learning a first language is somehow a matter of ‘training’ is preposterous on the face (p.217) of it (not to say strikingly lacking in detail). Who trained you, for example?

  • 5. There is no reason to believe that the content of a mental representation (that is, its referent) has anything to do with its role in inference (insofar as its role in inference is identified with its role in mental processes; that is, with its role in mental computations). There is likewise no reason to believe that the content of a mental representation has anything much to do with its causal relation to other mental representations; all that counts is its causal connections to the world. A fortiori, there is no reason to believe that the metaphysics of meaning imposes a ‘charity’ condition, or a ‘rationality’ condition upon ascriptions of beliefs.

  • 6. There is no reason to believe that the content of one's concepts is determined by (e.g.) the character of one's behavioral capacities. Pragmatism is false. It does remain open, however, that content‐determining mind‐world causal connections are mediated (sometimes or always) by the exercise of such capacities. If so, then what actions one can perform is (contingently but not metaphysically) relevant to the content of one's concepts.27

  • 7. There is no particular reason to suppose that cognitive development comes in ‘stages’. So far as I know, all the classical versions (including, most notably, Piaget's) depend on assuming that the content of mental representations is what changes from stage to stage, and that content supervenes on inferential role. Neither of these is true according to the present treatment.

  • (p.218)
  • 8. There is no reason to suppose that ‘how you think’ or ‘what you can think about’ depends on what language you speak. Nothing but the semantics of Mentalese determines what one can think, or think about, and the semantics of Mentalese is prior to the semantics of English.

  • 9. There is no reason to suppose (as, however, many philosophers appear to do) that demonstration presupposes conceptualization; for example, that you can only demonstrate something if you have a concept that subsumes it (or, perhaps, that you believe subsumes it). In fact, the whole thrust of the discussion has been that the priorities go the other way around: relations depend on causal relations, not vice versa. This is, of course, essential if any reductionist program along the lines of PT is to get off the ground.

  • 10. In fact, the story I've been telling you comports rather nicely with the story that Zenon Pylyshyn likes to tell about FINSTS. FINSTS are preconceptual pure demonstratives of Mentalese. The visual system uses them to effect a primitive sort of reference; in particular, to pick out distal objects in the course of very early28 visual processes. The way FINSTS work is: they get ‘grabbed’ (in circumstances that are presumed to be specifiable in purely psychophysical terms and thus do not presuppose conceptualization)29 by ambient things and events, which they then ‘stick to’ in ways that frequently tolerate changes in the target's location, speed, trajectory, color, apparent shape, and even its disappearance due to visual occlusions. Since you can, if Pylyshyn is right, achieve this primitive sort of reference without the application of concepts to the (p.219) referent, the idea that there is an internal connection between referring and conceptualizing would turn out not to be true. I don't know whether Pylyshyn's story about how pure demonstration works in early vision is even approximately true, but it's certainly very attractive; and I would be vastly surprised if it could be refuted by an appeal to a priori constraints on the relation between reference and conceptualization.30

  • 11. There may be no good reason for supposing that English has a semantics at all; perhaps the only thing that does is Mentalese. If that's right, then what are usually called ‘semantic level’ representations of English sentences (or representations of their ‘logical form’) are really no such things. Rather, they should be taken as representing translations of English into Mentalese. The translation of a sentence in Mentalese is, of course, no more a representation of that sentence than is its translation in French. To suppose otherwise would be to commit what in the old days they called a ‘category mistake’.

  • The implications for linguistic theory (if not for linguistic practice) might be quite interesting. For example, it could turn out that there is no such thing as the ‘interface’ between semantic‐level representations and syntactic‐level representations (because there is no such thing as semantic‐level representation). We might also want to reconsider the conventional view, embraced for expository convenience in the preceding chapters, that the content of English is compositional. (By contrast, the content of Mentalese representations had better be for all the familiar reasons: productivity, systematicity, and so forth.) If it turns out that Mentalese is compositional and English isn't, then, very likely, not every Mentalese sentence has a translation in English. That's (p.220) perfectly fine. Who but a philosopher of the pragmatist persuasion would claim that you can say whatever you can think?31

And so forth. I'd be very surprised if there aren't many other things that most philosophers and cognitive scientists believe but that a serious, naturalizable, computational version of RTM would refute.

SNARK. If only we had one. AUTHOR. If only we had one.


(1) Basic physics is, of course, a moving target; our estimate of what the material world is made of has changed a lot since Lucretius proposed atoms and the void. No doubt it will change still more before we're finished. That doesn't, however, make the physicalist thesis vacuous, nor does it write a blank check on the ontology of future science. For example, PT implies that there is a unique basic science (in effect that the special sciences are arranged in some sort of hierarchy, a claim for which plausible evidence is already in hand) and it requires that how the ontology of basic science turns out determines the ontology of all the rest of science. None of that is truistic; maybe none of it is even true.

(2) I guess what PT really is, is synthetic a priori: In principle it's subject to change under empirical pressure, but de facto it's universally presupposed in empirical theory construction.

(3) Come to think of it, it isn't entirely clear that any middle‐sized objects (mountains or typhoons, to say nothing of cabbages or kings) really are constituted entirely by things of the sort that the ontology of basic physics is prepared to recognize. But intentional psychology is especially worrisome since (presumably) none of the other special sciences invokes semantic/intentional properties as part of its explanatory apparatus. Presumably we don't need notions of truth, reference, content, and the like to tell the scientific story about mountains, typhoons, or even cabbages. Kings, however, seem to be a different kettle of fish. In short, it's conceivable that everything except intentional/semantic states and processes is material. Naturalism about psychology sets its face against this possibility.

(4) Or, perhaps, in speech acts that such expressions are used to perform; the distinction doesn't matter for present purposes.

(5) Or perhaps geology is also a façon de parler and there is actually no science except basic physics. People have been known to say that. People say the oddest things.

(6) This leaves open as the semantics of ‘logical’ expressions: connectives, quantifiers, and so forth. For what it's worth, I'm inclined to think that ‘and’ is defined by its truth‐table (and not, for example, by its ‘inferential role’). I don't think this view commits a circularity, but I won't try to convince you of that.

(7) Assuming this helps with PT of course; but there's another, more parochial reason why it would be nice to have a causal account of the semantics of LOT. It's an argument that's been raised against the LOT picture very many times (indeed, against RTM as such) that Mentalese formulas themselves require an internal interpreter if they're to be meaningful. The immediate implication would be a homuncular regress, which is not a pretty thought. This line of argument has proved remarkably resilient over the last couple of decades; it has survived repeated decapitation. Patently, however, it is defanged if the content of a mental representation is determined not by the results of its interpretation but by its causal connections to things in the world.

(8) For this and related notions see Soames (2002).


SNARK. That's rather a lot, don't you think. AUTHOR. I'd make more if I could think of any that might help. Since I have to start somewhere, it might as well be somewhere close to where I'm going.

(10) The resemblance, though, is pretty superficial. I'm not claiming that the inside‐the‐perceptual‐circle vocabulary is rich enough to define the outside‐the‐perceptual‐circle vocabulary, nor am I claiming that it is restricted to terms that express ‘sensory’ properties (whatever exactly those are), or that it is in any way epistemologically privileged. I don't believe any of that; nor should you. In‐the‐circle vocabulary is just the sort of stuff you learn from Mummy when she says, as it might be, ‘That's a duck, dear’. Still, if the things I've been saying are right, there's something to the empiricist view that facts about perception have a special role to play in concept acquisition. Let us all praise famous persons.

(11) This kind of example is especially exasperating if you're the sort of philosopher who believes that charity is constitutive of interpretation. Since, in the circumstances imagined, interpreting John as referring to a cow‐or‐cat makes more of his thoughts true than interpreting him as referring to a cat, it appears that the charitable interpretation will be the wrong one. There is a close connection between the claim that no causal theory of reference can cope with the wild‐tokens problem and the claim that no causal theory of reference could offer an adequate account of false assertion or false belief.

(12) This is not, perhaps, the way that Davidson intended to be read; I have heard him say that he had no truck with transcendental arguments. So either my parsing of his texts is wrong or Davidson was wrong about what he had any truck with. The present considerations don't depend on choosing.

(13) In particular, Davidson's radical linguist can't make any very rich assumptions about the content of the native informant's propositional attitudes, since, for Davidson, the problem of the radical interpretation of an informant's language is essentially identical to the problem of identifying the content of his beliefs, desires, and so forth.

(14) Perhaps I should repeat that I believe no such thing. What I'm doing is to sketch the galaxy of assumptions from which Davidson's story about triangulation seems to have emerged.


SNARK. Snakes have hair? AUTHOR. This is a philosopher's snake. A philosopher's snake has hair if the philosopher whose snake it is says that it does.

(16) It seems Davidson doesn't think a problem of radical interpretation with respect to one's own utterances. It is unclear to me that he has any right not to think that.

(17) It is, however, an empirical question whether interpreters really are able to draw that sort of inference in those sorts of circumstances with any significant reliability; and, in fact, the evidence is that they can't. S's are surprisingly bad at guessing what a speaker might be saying even when it's plausible that the speaker is saying something about the situation that he and the interpreter share; this is especially true for the interpretation of verbs (see Gleitman et al. 2005).

(18) Hence the philosophical joke: ‘Q. How many philosophers does it take to see a tree? A. Infinitely many: one philosopher to look at the tree, another one to interpret the one who looks at the tree, and still another one to interpret the one who interprets the one who looks at the tree, etc. I don't suppose you found that joke hilariously funny. Philosophical jokes generally aren't.

(19) It seems Davidson also thinks that the significance of triangulation for interpretation provides a transcendental argument for the essentially ‘social’ character of language: a sort of ‘private language’ argument after the fact. I'm not going to consider this proposal, since it's patently incompatible with the computational account of cognition to which this book is for better or worse committed.

(20) Davidson offers, as far as I know, no fully worked‐out theory of triangulation. The quotations in the text are culled from a scattering of his essays reprinted in 2001, to which my citations refer.

(21) I like counterfactuals, but many philosophers doubt they can bear much weight. I don't think, however, that Davidson could himself have avoided resting a lot upon them. Many thoughts don't actually get uttered; and many utterances don't actually get interpreted. There is no Great Interpreter in the sky to do for their intentional contents what God did for Berkeley's tables and chairs; namely, sustain them when nobody else is around to do so. Counterfactuals look like being the only candidates for an ontology of interpretation. Some of the implications of this are far afield and very surprising (see Fodor 2008).

(22) Or a ‘counterpart’ of Adam himself if you prefer that way of doing business.

(23) I propose not to worry about the ontology of reference; e.g. about whether Adam's reference is to an event, or to an object, or to both, or to neither. I think, in fact, that answering that question requires more information than my story about Adam has thus far supposed his interpreter to have. But never mind. (Or, if you do mind, see Fodor 1994 for some discussion.)

(24) This helps, since, of course, what you actually say (like what you are actually disposed to say) depends not just on the content of your perceptual belief, but on how that belief interacts with a background of your other beliefs, desires, and so forth. This is one of the ways that framing the issue as the interpretation of a speaker's utterance (rather than the interpretation of the speaker's thoughts) drags you into holism. I take it for granted that holism is a place to which it is good not to be dragged.

(25) It perhaps bears repetition that the definite descriptions in question needn't provide analyses of the representations they introduce; indeed, the representations they introduce may perfectly well be primitive, hence not analyzable. So, the distinction between representations whose referents are fixed by triangulation and representations whose referents are introduced by description need not be principled and, in particular, may have no particular relevance to questions about the individuation of the representations.

(26) Using a language for communication requires implementing some sort of procedure that its speaker/hearers can use for translating it into and out of the language of thought. But it doesn't follow that there must be such a procedure in order for there to be a language of thought.

(27) But there's a caveat. It remains open that although there are no rationality (etc.) constraints on what contents a propositional attitude can have, there may well be such constraints on a propositional attitude with that content being a belief. Believing may be functionally defined even though believing that P is not. You need two parameters to specify an attitude: roughly, believing that P versus desiring(/preferring, intending, etc.) that P and believing that P versus believing that Q. It has never been very clear how conditions of charity, rationality, and the like would apply in the case of attitudes that don't have truth‐values.

(28) Computationally early, not (or not necessarily) ontogenetically early. Still less phylogenetically early.

(29) Somebody lights a cigarette in a dark room; all eyes move reflexively to foveate the light and a FINST is reflexively assigned. No conceptualization of the stimulus is required to mediate the process; not even conceptualization of the stimulus as a light. In this sort of case the direction of causation is, as it were, from outside in, not from inside out; so it's the reverse of what I suppose happens when a concept is assigned to a percept.

(30) Still less by appeal to a priori constraints on the relation between demonstration and consciousness. Compare Campbell (2002).

(31) Still less plausibly that it's a priori that you can say whatever you can think; that would imply, for example, that animals can't think anything at all. Mr James, who is the domestic feline currently in residence here, takes the slight to heart.

‘But animals can't think in the same sense of the term that we do.’ ‘Well, in what sense of the term do we think? Presumably it's the sense of “think” in which “think” means think. Surely, if animals think at all, they too must think in that sense of the term. How else?’