Unpacking a dog

Jerry Fodor

  • A Study of Concepts by Christopher Peacocke
    MIT, 266 pp, £24.95, December 1992, ISBN 0 262 16133 8

The Modern era, as analytic philosophers reckon, started with Descartes. By contrast, the Recent era started when philosophy, in Richard Rorty’s phrase, took the ‘linguistic turn’. So it started with Frege or Russell, or early Wittgenstein, or the Vienna Circle; take your pick. Modern philosophy was mostly about epistemology: it wanted to understand what makes knowledge possible. Recent philosophy is mostly about meaning (or ‘content’) and wants to understand what makes thought and language possible. So, anyhow, we tell our undergraduates when we’re in a hurry.

There’s something to it, but probably not much. ‘Transcendental’ arguments used to run: ‘If it weren’t that P, we couldn’t know that Q; and we do know that Q; therefore P.’ Philosophical fashion now prefers: ‘If it weren’t that P, we couldn’t say (or think or judge) that Q; but we do say (or think or judge) that Q; therefore P.’ Much of a muchness, really. The two kinds of arguments tend to be about equally unconvincing and for the same reasons Often enough, it’s Empiricist preconceptions that do the work in both.

This is not, however, to deny that there is something very peculiar about Recent philosophy. There has indeed been a change that goes much deeper than shifting styles of philosophical analysis. What’s really happened, not just in philosophy, but in psychology, lexicography, linguistics, artificial intelligence, literary theory and just about everywhere else where meaning and content are the names of the game, is a new consensus about what concepts are. Take a sample of current and Recent theorists, chosen with an eye to having as little in common as may be: Heidegger, or Wittgenstein, or Chomsky, or Piaget, or Saussure, or Dewey or any cognitive scientist you like, to say nothing of such contemporary philosophers as Davidson, Dennett, Rorty and Quine. You may choose practically at random, but they are all likely to agree on this: concepts are capacities; in particular, concepts are epistemic capacities, abilities to recognise or to infer.

Christopher Peacocke’s A Study of Concepts is about as subtle and sophisticated an elaboration of the idea that concepts are epistemic capacities as you will ever want to read. It may, in fact, be a more subtle and sophisticated elaboration of that idea than you will ever want to read. Peacocke is hard work and he spares his reader nothing. His prose is not, perhaps, denser than the intricacy of his thought requires, so I’m warning, not com plaining; but his book wants exegesis, and it will surely get a lot. Many’s the graduate seminar that will slog its way through, line by line, and will be edified by doing so.

I won’t attempt anything of that sort here. There are too many passages I do not understand and, of the ones I do understand, there are too many that I haven’t made up my mind about. It does seem to me, however, that a striking number of Peacocke’s moves depend on assumptions that he makes, explicitly but practically without argument, in the book’s first several pages – and which strike me as a symptom of our times.

Peacocke’s topic is the nature of concepts. Just roughly and by way of orientation: 1. Concepts are word meanings. The concept dog is what the word ‘dog’ and its synonyms and translations express. This ties theories of concepts to theories of language. 2. Concepts are constituents of thoughts. To think that dogs bark is inter alia to entertain the concept dog and the concept bark. 3. Concepts apply to things in the world. The concept dog is one which, of necessity, all and only dogs fall under. Judgments are applications of concepts, which is why it’s things in the world that make judgments true or false.

This catalogue is me, not Peacocke, but I don’t expect it’s anything he’d object to very much. So then, if that’s what concepts are, what should a theory of concepts be? Starting on page five: Throughout this book I will try to respect the following principle ... There can be nothing more to the nature of a concept than is determined by ... a correct account of “grasping the concept” ... a theory of concepts should be a theory of concept possession.’ There are, to be sure, trivialising readings of this equation (C is the unique concept whose possession condition is that you have the concept C). But Peacocke intends that the nature of a concept should be illuminated by what a theory says about grasping it. For example: ‘Conjunction is that concept C to possess which a thinker must find [inferences of certain specified forms] primitively compelling, and must do so because they are of these forms.’ It partially identifies C as the concept of conjunction that anybody who has it finds inferences from the premises p and q to the conclusion pCq compelling as such.

Peacocke is saying that sometimes part of what grasping a concept comes to is being able to see, straight off that certain of the inferences that it plays a part in are okay. Another part of what grasping a concept sometimes comes to is being able to see straight off that the concept applies to something one perceives. In either case, though Peacocke’s question is nominally semantic – It’s about conceptual and/or linguistic content – epistemology is calling the shots. What concepts you have depends on what concepts you grasp, and what concepts you grasp depends on what epistemic capacities you have. Having conjunction, for example, is being able to see certain inferences as valid. It is arguable that the linguistic turn from epistemology to semantics doesn’t amount to much when meaning is itself epistemically construed.

Both the idea that theories of concepts are theories of concept possession and the idea that possessing a concept is having certain epistemic capacities tend, these days, to be treated as truisms. In fact, they are intensely tendentious. I suppose Descartes, or Hume, or Mill would have said that you identify a concept not by saying what it is to grasp it but by saying what it is the concept of. Accordingly, on this older view, to illuminate ‘the nature of a concept,’ you need, not ‘a theory of concept possession’, but a theory of representation. The key question about, as it might be, the concept dog is something like: ‘In virtue of what does that concept represent dogs and in virtue of what do other concepts fail to do so?’

Changing the topic from ‘How do concepts represent?’ to ‘What capacities constitute a concept?’ was, I think, what really started Recent philosophy. And arguably Dewey and the Pragmatists had more to do with it than Frege or Wittgenstein. It’s a paradigmatically Pragmatist idea that having a concept consists in being able to do something. By contrast, uninstructed intuition suggests that having a concept consists in being able to think something. (Having the concept dog is being able to think about dogs; or, better, about the property of being a dog.) In my view, uninstructed intuition has the stick by the right end.

So that joins the issue: Is a theory of concepts a theory of concept possession or is it a theory of how concepts represent? ‘Why not both?’ you might ask, in one of your ecumenical moods. But this would be to miss the metaphysical tone of Peacocke’s demand for a theory of concepts. No doubt there might be both a theory of what it is to grasp a concept and a theory of how a concept represents; and there might be theories of how a concept is acquired and how it’s applied too, while we’re at it. But none of these would be a conceptual analysis unless it specifies the properties that make a concept the concept that it is. Peacocke is saying that what makes a concept the concept that it is are the conditions for possessing it.

Because he thinks philosophical analyses unpack the possession conditions that you need to individuate concepts, he is prepared to take quite a hard line on the methodological priority of philosophical investigations in the cognitive sciences

An agenda for psychology suggested by the general approach I have been advocating is, then, this: For each type of thinker and for each concept possessed by a thinker of that type, to provide a sub-personal explanation of why the thinker meets the possession condition for that concept ... Carrying out this agenda is also in its very nature an interdisciplinary enterprise. For any particular concept, the task tor the psychologist is not fully formulated until the philosopher has supplied an adequate possession condition for it.

The idea that philosophy sets the agenda for psychology, or for any other empirical inquiry, is a typical product of the idea that philosophy is conceptual analysis and that conceptual analysis is, as Peacocke likes to say, ‘relatively a priori’. It strikes me, frankly, as ahistorical and maybe a touch hubristic. The susurration that you hear is legions of cognitive psychologists not holding their breath till their task is fully formulated by philosophers.

I doubt that the theory of concepts which engenders this account of analysis can be sustained. The key problem is that people who have the concept dog thereby have all sorts of capacities that people who don’t have that concept fail to have. And, surely, not all these cap acities are essential to ‘the nature of the concept’. If I didn’t have the concept dog, I suppose I couldn’t have the concepts dog bath or dog bone Or fighting like cats and dogs. And, if I didn’t have those concepts, there would be all sorts of inferences that I would fail to find ‘primitively compelling’ (from ‘dog bath’ to ‘bath for dogs’, for example) and all sorts of perceptual judgments that I would be unable to make (that the object currently on display is a dog bath rather than a bird bath, for example). But I suppose that none of these capacities illuminates the essential nature of the concept dog; one could have the concept even if one had none of them.

If there is a capacity, to possess which is constitutive of grasping the concept dog, and if it consists, inter alia, in being able to see certain inferences as primitively compelling, then there must be something that distinguishes these individuating inferences from the rest. And, as we’ve seen, it has to be something non-trivial if conceptual analysis is to be worth doing. (It’s constitutive but trivial that dog is a concept that lets you make inferences about dogs. It’s non-trivial, but also non-constitutive, that dog is a concept that lets you make inferences about dog baths.) Are there some inferences that are both non-trivial and constitutive of the concept dog? The answer must be yes, if theories of concepts are theories of their possession conditions and possession conditions are inferential capacities.

Very well then, what determines which in ferential capacities are non-trivially constitutive of grasping the concept C? Which of the cluster of capacities that grasping C may bring in train are the ones that belong to its possession conditions? Peacocke says a lot about which possession conditions are constitutive of one or other concepts, but remarkably little about the general questions. The closest we get is this: ‘Distinctness of concepts. Concepts C and D are distinct if and only if there are two complete propositional contents that differ at most in that one contains C substituted in one or more places for D, and one of which is potentially informative while the other is not.’ So, for example, the concept dog is distinct from the concept barker because someone who has fully grasped the former concept, and who takes it that dogs are animals, might nevertheless take it to be news that barkers are animals. Whereas (assuming that dog is the same concept as domestic canine) nobody who takes it that dogs are animals could find it news that domestic canines are. Or, to put the same idea in terms of possession conditions: finding the inference ‘if dogs are animals then domestic canines are animals’ primitively compelling is among the possession conditions for dog. Whereas, though the inference ‘if dogs are animals then barkers are animals’ is quite a good inference, finding it primitively compelling would presumably not count as constitutive for any of the concepts involved. This would be true even if, in point of fact, all and only dogs bark. (I emphasise that the example is mine, and that it is crude.)

The notion of an informative proposition (/inference) thus looms large in Peacocke’s treatment. He needs it very much if he is to avoid the trivialisation of his project. As far as I can tell, Peacocke thinks it’s too obvious to bother arguing for that you can individuate possession conditions, and thereby flesh out the notion of concept identity, by appealing to the informativeness test. I don’t at all share his optimism. Someone who finds it unsurprising that John understands that bachelors are bachelors might, I suppose, still wonder whether John understands that bachelors are unmarried men. So it appears that if, following Peacocke’s recipe, you substitute ‘unmarried men’ for the second ‘bachelors’ in ‘John un derstands that bachelors are bachelors’ you go from something unsurprising to something that someone might find news. Since, however, the concepts bachelor and unmarried man are identical if any concepts are, it looks like the informativeness test for concept identity is badly undermined. These so-called ‘Mates cases’ (after their inventor, Benson Mates) are a philosophical commonplace; Peacocke doesn’t mention them, but I don’t understand why they don’t worry him

Or consider poor Jones, who went off with a bang: he didn’t know that being flammable and being inflammable are the same thing. Jones would have found it informative had someone taken the trouble to tell him; if someone had, he would be with us still. Yet ‘flammable’ and ‘inflammable’ are synonyms and hence must express the same concept if concepts are word meanings. It looks, again, as though informativeness is one thing, conceptual identity another. Peacocke has a remark arising in a quite different context, that may be intended to cover this sort of case. ‘In this particular example, it suffices for a theory of concepts to aim to explain those patterns of epistemic possibility that exist only for one who fully understands [the corresponding word] (and any synonyms he may acquire.)’ (Peacocke’s parentheses, my emphasis.) I doubt, however, that Peacocke intends this as a codicil to his informativeness test, since it presupposes a notion of synonymy which is itself semantic and, to put it mildly, unexplicated.

I am, truly, not meaning to quibble or to insist on what are arguably marginal counter-examples. Like lots of other philosophers who have been influenced by Quine, I really do doubt that concept identity can be explicated unquestion-beggingly by appeal to notions like informativeness. I doubt, in fact, that it can be unquestion-beggingly explicated at all so long as you think of having a concept in terms of possessing diagnostic epistemic capacities. (Whether concept identity can be explicated unquestion-beggingly in non-epistemic terms is a long question to which the short answer is ‘maybe’.) My own guess is that there aren’t any non-trivial inferences that the concept dog requires its possessors to have come what may. As Putnam has pointed out, even dogs are animals would fail in science-fiction worlds where dogs turn out to be Martian robots. In such a world, somebody could ‘fully grasp’ the concept dog but find the inference dog→animal uncompelling: indeed, unsound. Nor, I think, are there any perceptual judgments which dog owners as such must be able to make. No landscape is so uncluttered that it is impossible in principle that one should fail to recognise that it contains a dog.

The problem is that Peacocke’s whole project, his whole conception of what concepts are, and hence of what a theory of concepts should aim for, is committed to an epistemic distinction between analytic (constitutive) capacities and synthetic (collateral) capacities. And this is a distinction that Peacocke, like the rest of us, doesn’t know how to draw. (He takes note of this commitment in a passing footnote and is unperturbed by it.) This leads to a geographical impasse: if, as people on my side of the Atlantic are increasingly inclined to suppose, there isn’t an epistemic analytic/synthetic distinction, then the notion of a possession condition is infirm and you cannot identity grasping a concept with being disposed to draw the inferences by which its possession conditions are constituted. Correspondingly, philosophical analysis can’t consist in setting out the possession conditions that concepts have.

The long and short is: I think there is good reason to doubt that the kind of philosophy Peacocke wants to do can be done. In one passage, he remarks, almost plaintively, that ‘theories are developing in the literature of what it is to possess certain specific concepts ... While there is much that is still not understood and not all of what has been said is right, it is hard to accept that the goal of this work is completely misconceived.’ I guess I don’t find it all that hard. The linguistic turn was, I think, an uncompleted revolution; really to turn from theories of knowledge to theories of meaning, you would have to stop construing content in epistemological terms. Many analytic philosophers can’t bear not to construe content in epistemological terms because they think of philosophy as conceptual analysis and they think of conceptual analysis as displaying the epistemic capacities that grasping a concept requires. If, as I believe, that whole picture is wrong, a certain kind of analytic philosophy is ripe for going out of business. If there is no analytic/synthetic distinction, then there are no analyses. This is a thought that keeps philosophers on my side of the Atlantic awake at night. Why doesn’t it worry more philosophers on Peacocke’s side?

Does any of this really matter except to philosophers over sherry? Oddly enough, I think perhaps it does. We are in the midst of a major interdisciplinary attempt to understand the mental process by which human behaviour accommodates to the world’s demands: an attempt to understand human rationality, in short. Concepts are the pivot that this project turns on since they are what mediate between the mind and the world. Concepts connect with the world by representing it, and they connect with the mind by being the constituents of beliefs. If you get it wrong about what concepts are, almost certainly you will get the rest wrong too.

The cognitive scientists I know are mostly a rowdy and irreverent lot, and I shouldn’t want to be around when they hear Peacocke’s views about the primacy of philosophy in defining then enterprise. But it’s perfectly true that they have, almost without exception, assumed what is essentially a philosophical theory of concepts, and that it’s pretty much the one that Peacocke also takes as given: concepts are epistemic capacities. In consequence, questions about which capacities constitute which concepts perplex the whole discipline, and nobody knows how to answer them. If it turns out that concepts aren’t capacities, these question don’t have answers.

I’m not proposing a transatlantic methodological shoot-out, but I do think there needs to be a sustained discussion of what concepts are, and I think that de-epistemologising semantics – completing the linguistic turn – is likely to be its outcome. If so, theories of language and mind will come to look very different from what Recent philosophy has supposed; and the project of philosophical analysis will look inconceivably different. (Assuming that there is any such project left.) In this discussion, someone will have to speak with insight and authority for the Received View. Peacocke has done that, and we are all in his debt.