Doing Without Whatís Within; Fiona Cowieís Critique of Nativism

by Jerry Fodor

Rutgers University

PART 3: The Impossibility Arguments.

        Iím afraid I am now required to set out some background. I must trace the course of an argument Iíve been having (mostly with myself) for the last twenty five years or so, as to whether, given plausible empirical premises, it is even coherent to hold that there is such a process as concept learning. And, if itís not, what nativistic alternatives there might be.
        I managed some time back to convince myself that Impossibility Arguments show that the received account of concept learning is indeed incoherent. But it has recently occurred to me that the implications of this can perhaps be made to sound a little less preposterous than, for example, that the concept CARBURATOR (or the concept CURRY) is innate. Contrary to what I had at first supposed, there is a way of saying things like `no concepts are learned; a fortiori, the concept CARBURATOR isnít learnedí that makes it not also require saying things like `no concepts are learned; a fortiori the concept CARBURATOR is innate.í
        That strikes me as, if perhaps not awfully important in the long run, still a good thing tactically. For some reason, philosophers, who are often prepared to swallow the most outlandish views ----that there arenít any tables or chairs; or that there arenít any numbers; or that there arenít any minds; or that, (to the contrary) there is
nothing but minds; or that there is nothing but numbers; (I havenít heard of a philosopher according to whom there is nothing but tables and chairs, but my knowledge of the ontological literature is fragmentary); or that we made the stars; or that there is no distinction between confirmation and truth; or that the only good is the greatest happiness of the greatest number; or that the goal of physics is to predict the state of excitation of oneís sensory neuronsÖ and so forth, practically endlessly--- philosophers, who have learned to gaze on all of that and not to boggle, tend to become quite hysterical at the thought that the human conceptual repertoire, CARBURATOR included, might be innately specified. Their view, apparently, is that human ethology (unlike, say, spider ethology, or fish ethology) is an a priori science, primarily responsible to what strikes philosophers as plausible from a genetic or an evolutionary point of view.
        That being so, and what with the notion of concept learning being incoherent (according to an argument I find convincing) it would be nice if one could somehow endorse the Impossibility Argument without having to say that CARBURATOR is innate
        Cowie, of course, thinks that IAs are unsound, hence that we neednít worry about what form of concept nativism the philosophical community might be prepared to tolerate. Partly she thinks this on methodological grounds, but mostly on the ground that a key premise of IA isnít true. As it turns out, Iím comprehensively unmoved by the considerations she raises; Iíll tell you why in just a moment. First Iíll have to give you a sketch of how IA is supposed to run. Then Iíll tell you what kind of concept nativism I think we ought to endorse if IA is sound; and why I think that kind of concept nativism is independently plausible. Then Iíll tell you why Cowie rejects (not just the Impossibility Argument but, also and independently,) the kind of concept nativism Iím proposing. Then I will tell you why her grounds for rejecting it are insubstantial. Then, I think, we can call it a day.
        IA runs on the following assumptions, from which, it claims, the incoherence of the received view of concept learning follows:
3.1. Environmentally caused alterations of a creatureís conceptual repertoire count as concept learning only if they are mediated by processes of hypothesis formation and confirmation; if one were somehow to acquire the concept DOORKNOB by surgical insertion, that would not count as learning it. I take it that this is actually not in dispute. Surgical insertion is not a species of hypothesis formation; and to my knowledge, no alternative to the hypothesis testing account of concept learning has ever been proposed. (There are, to be sure, many different vocabularies that this hypothesis has been couched in, and people who espouse it thus often fail to notice that theyíve done so.)
        It thus bears emphasis that, if you accept 3.1, you
already have good reason to doubt that the notion of concept learning is coherent. What hypothesis confirmation eventuates in confirming is, after all hypotheses; and concepts are not hypotheses (a muddled Pragmatist tradition to the contrary notwithstanding). You can, for example, (dis)confirm the hypothesis that dogs bark; but you canít (dis)confirm the concept DOG or the concept BARK. That concepts arenít hypotheses should hardly seem surprising since concepts are the constituents of hypotheses; concepts are what hypotheses are made of and are thus prior to hypotheses, in much the ways that bricks are prior to brick houses. Since concepts are prior to hypotheses, they are a fortiori, concepts are prior to the (dis)confirmation of hypotheses. Empiricists have been confused about these priority relations between (what used to be called) `Ideasí and `Judgementsí for several centuries, and the end of this also is not in sight. Just as Kant and Frege both warned it would, confusing Ideas with Judgements got empiricists into endless trouble, including their egregious failure to understand that theories of concept acquisition must differ in kind from theories of belief fixation; in particular, that the former canít be learning theories as 3.1 understand that notion.28
3.2. `Mostí of our concepts are primitives; i.e. they have no internal structure; i.e. they havenít got other concepts as constituents (in the way that, for example, itís often supposed that the concept DOG has the concept ANIMAL as one of its constituents, and that the concept BACHELOR has the concept UNMARRIED as one of its constituents.)
        Whereas 3.1 is more or less untendentious (presumably because its unsettling implications have not been widely recognized), 3.2 very clearly isnít. I take it, however, that 3.2 is licensed by the following subsidiary argument:
        3.2.1 Not
all concepts could have other concepts as parts (at least some concepts must be `primitiveí). This I take to be common ground.
        3.2.2 Broadly empirical considerations (from cognitive psychology and elsewhere) show that `mostí concepts could have internal structure only if most concepts are (something like) stereotypes or prototypes. (For discussion of some of this literature, see Fodor (
        3.2.3 There are decisive reasons why `mostí concepts canít be (anything like) stereotypes or prototypes.
        Cowie has two main lines of attack on the soundness of the Impossibility Argument, one of which centers on 3.2.3; weíll turn to that presently.
        3.2.4 Primitive (unstructured) concepts canít be learned by the formation/ confirmation of hypothesis.
        The basic argument for 3.2.4. is that its denial leads to circularity. Consider a concept like RED (which is pretty widely agreed to be primitive if any concept is.) How would a hypothesis testing account imagine that RED is learned? Well, presumably learning RED would involve confirming some hypothesis about
which concept RED is (about its `individuating propertiesí); as, for example, that itís the concept that expresses the property of being red. But, clearly, that canít be right; for any hypothesis of the form X is the concept that expresses the property of being red ipso facto contains the concept RED among its constituents. A fortiori, itís not a hypothesis that could be formulated by someone who lacked that concept. A fortiori itís not a hypothesis that could be (dis)confirmed by anyone who lacked that concept. Since the same reasoning goes through, mutatis mutandis, for any concept that is supposed to be primitive, it follows that primitive concepts canít be learned (where learning means what 3.1 says that it does). Hume got this right: "Öwhenever we reason, we must antecedently [my emphasis] be possest of clear ideas, which may be the objects of our reasoning. The conception always precedes the understandingÖ (214)." You canít reason with a concept you donít already have.30 So `mostí primitive concepts canít be acquired by reasoning. But, taken together, 3.2.2 and 3.2.3 imply that `mostí of our concepts are primitive. So `mostí of our concepts are unlearned.
        Just a word about the shudder quotes around `mostí. Clearly we have infinitely many structured concepts (ones that have other concepts as constituents) Thus my concept A FRIEND OF MY AUNT contains, among its parts the concepts FRIEND and AUNT, as do infinitely many concepts that belong to the same family: A FRIEND OF A FRIEND OF MY AUNTÖ and so on. This is, once again common ground. But what of the concepts FRIEND and AUNT themselves? Are they primitive, or do they have parts? And if the latter, what parts do they have? Itís clear, in any case, that if AUNT has constituents, the corresponding English expression (viz. `Auntí) doesnít display them; (unlike, of course, the English expression that corresponds to A FRIEND OF MY AUNT, (viz ` a friend of my Auntí) which, as it were, shows that FRIEND and AUNT are parts of the complex concept that it expresses.) All that being so, we can now take the quotes off `mostí. The conclusion of the Impossibility Argument is supposed to be that the set of unlearned concepts is approximately coextensive with the set of concepts whose structures are
not displayed by the corresponding English expressions. (It therefore likely includes FRIEND and AUNT, but not FRIEND OF MY AUNT.) This is all very approximate, to be sure, but it will do for the purposes of hand since I suppose that, if anything of even approximately this sort is true, then empiricism isnít.
        So now, at last:

3.3 Cowieís objections to the impossibility argument.

        In an earlier draft of this paper, I allowed myself a little grumble about Cowieís tendency to offer, against some proposition she has under attack, a fardel of arguments the conjunction of whose premises is not consistent and some of which must therefore be unsound.31 Itís hard on the weary exegete that Cowie generally doesnít say which arguments she proposes to give up in case she canít have them all. In the event, however, I decided to delete that passage. (I think it is good for my character occasionally to resist the temptation to grumble. Very occasionally.) But the reader should be advised that weíve now come to a polemical situation of this kind.
        Cowie has two objections to impossibility arguments. One is that (pace 3.2.2) most concepts are prototypes, and itís common ground that prototypes are complex statistical structures and can be learned by assembling them from their constituents. The second argument, however, takes a much stronger line in that it seems to reject, a priori, the very idea that a concept might be innate. Now, I really donít think Cowie can have this both ways. Prototypes are ipso facto constructions out of a primitive conceptual basis; and, as far as I can tell, Cowie accepts that primitive concepts have to be unlearned (as per 3.2.4). But if that is so, she can hardly claim to be possessed of a general and principled argument that no concepts are innate.
        In short: Cowieís empiricist account of how concepts are acquired applies
only to concepts that have (or are) prototypes. But if prototypes are ipso facto learnable, thatís because they are ipso facto structurally complex, hence not primitive concepts. So, it looks to me that, qua friend of prototypes, Cowie needs there to be a bona fide set of innate primitives. So, conceive of my puzzlement upon encountering such passages as this: "Fodor talks of [innate, primitive] concepts `becoming available,í as if acquisition were the activation by a triggering stimulus of some sort of preexisting conceptlike object. We come into the world equipped with a stock of `protoconcepts,í mental structures of some sort that become fully fledged concepts once they are triggered by an appropriate stimulusÖ Iíll argue that thisÖ picture is seriously confused. For thereís simply nothing for protoconcepts to be [sic] (83). " Well, I am confused; I donít see how it could both be that learned concepts are ipso facto complex and nonetheless that no concepts are innate. Thatís why not just paradigm rationalists, but also paradigm empiricists (Locke, Hume, William James) have always agreed that primitive concepts must be innate (for some references see Fodor, (1981)). How could a creature that has no concepts learn anything? (Say "bootstrap" and Iíll scream.)
        So, I donít think that both Cowieís objections to the impossibility argument could be sound; the positive (prototype based) view of concept acquisition towards which she gestures seems to me not to cohere with her claim that proto-(viz. innate) concepts are
ipso facto corrupt. I wonít, however speculate on how she might seek to reconcile these two sorts of argument. Since both of Cowieís objections to IAs are unsound, it doesnít matter, for our purpose, that their premises arenít compatible.

3.3.1. Cowieís argument against protoconcepts.

        (This is a short argument.) Innate concepts (like concepts that arenít innate; for that matter, like anything at all) are in want of principles of individuation. Now, patently, concepts are individuated by their contents inter alia; viz `semanticallyí. 32 Letís assume some or other sort of `externalistí metaphysics of content (eg. that the content of a concept supervenes on world-to-mind causal interactions.) Well, unactivated innate concepts ---those that are, as it were, waiting around to be triggered--- are presumably ipso facto not causally connected to anything in the world. So, according to externalism, they canít have any contents; so they canít be content-individuated; so they canít be concepts. "There is simply nothing for protoconcepts to be" compatible with, on the one hand, concepts being necessarily semantically individuated and, on the other hand, protoconcepts being de facto causally inert. 33
        So, the question comes down to: Could an externalist believe that there are innate ideas? Pace Cowie, the answer is: `Sure.í For example, an externalist could hold that the semantic properties of `protoconceptsí supervene on their
dispositions to enter into causal world-to-mind relations. Maybe what makes a mental representation a token of the protoconcept type CAT is its disposition to be triggered by cats. 34
        It is, I think, very puzzling that Cowie doesnít seriously consider the possibility of an externalist nativism that is dispositionalist about the semantic properties of concepts (all the more so since she does, briefly, consider the possibility of a dispositionalist
internalism; see the footnote before last.) Unless there are passages Iíve overlooked, the closest she gets is her remark (on p. 91) that "on one Ö model Ö experience serves to trigger innate protoconcepts, transforming preexisting mental objects Ö into fully fledged intentional objects. [However]Ö the assertion that protoconcepts are triggered by experience boils down to the observation, with which no one would disagree, that thereís something about our minds such that our experiences lead to our getting concepts." But no argument is provided that the former thesis does indeed `boil downí to the latter; and a momentís reflection suggests that it couldnít possibly. On all standard ethological accounts of triggering, part of whatís innate in a triggered concept is a specification of its proprietary trigger. Since the trigger of an innate concept is both proprietary and innately specified, such concepts can be unvacuously individuated by reference to what would trigger them; which is to say, by reference to their characteristic dispositions to enter into world-to-mind relations. 35
        Cowie thinks that postulating innate concepts should be avoided because it raises a pseudo-question to which no answer can be forthcoming: What constitutes the content of a concept when the concept is causally inert (eg. before it is triggered)? Iíve just argued, to the contrary, that the content of protoconcepts is no particular problem for a semantic externalist, so long as he assumes that it supervenes on (possibly unactualized) dispositions. But there is also a less narrow point to make; one that I think is sufficiently interesting as to merit (sigh, another) digression. The question about content that Cowie thinks that the postulation of innate concepts raises is of a kind that has familiar avatars outside nativist psychology. And itís one which, in consequence of the so-called `informational revolutioní in biology, we now have some idea how to answer.
        Put innate ideas to one side, and consider the structural similarity between two problems, the solution of each of which was crucial in determining the course of a science that raised it:

    Mendelís problem: What becomes of the properties of organisms when they arenít phenotypically expressed?
    (J.B.) Watsonís problem: What becomes of the intentional contents of propositional attitudes when they arenít the objects of thought?

        In both cases, there is the same crucial constraint on the answer. Unexpressed phenotypic properties neednít just `go awayí; they can skip generations and cause the offspring of heterozygotes to be more similar to their grandparents than they are to their parents. Likewise, the behavioral (etc.) expressions of oneís propositional attitudes are typically discontinuous; often, you can remember your name even across an interval of dreamless sleep. By contrast, however, causal chains canít skip links; they require that something going on all the time between the first component cause and the last component effect. So, whatís to do? How can it be that mental contents that arenít being thought, and phenotypic traits that arenít being instantiated, are nonetheless among the links in causal chains? These questions must have answers, whatever you may think about innate ideas and such.
        As indeed they do. Unexpressed traits (unattended contents) can be
coded for by microstructures that persist even through time stretches when the traits (/contents) donít manifest themselves. So, oneís `genes forí blue eyes can persist in oneís brown-eyed children, who may then themselves have children with `blue eyes just like Grannyísí. So too, the neural `engramí that encodes your knowledge of your name may continue to do so even while youíre asleep. Prima facie, these sorts of explanation of (what would otherwise appear to be) temporal gaps in causal histories are extremely persuasive. Watson himself went half bananas trying (and failing) to reconcile them with his behaviorism. (At one point, he was tempted by the thought that a sleeper who remembers that P is perhaps saying `Pí to himself, sotto voce, all through the night.) Mendel, being less methodologically inhibited, invented the gene.

3.4 Cowies argument for prototypes.

        Itís common ground that some such premise as 3.2.1 appears essentially in Impossibility Arguments 36 For, suppose that most concepts are prototypes after all. Then most concepts are complex, and could be learned by confirming hypotheses that identify the prototype. If the concept FISH is the prototype `wet, lives in the ocean and has scalesí then learning that fish are (typically) wet, scaly and ocean dwelling is all there need be to learning FISH. So sans an argument that most concepts arenít prototypes, IA fails.
        But there is such an argument, and itís short. Let C1 be a complex concept, of which the constituents are C2 Ö Cn (each of the latter may be either primitive or complex.) Then
: nothing belongs to the content of C1 except what belongs to the content of C2 or C3 orÖCn 37. Call this the Compositionality Constraint (=CC.) It says, in effect, that that the identity of a complex concept is entirely determined by the identity of its constituents. Since Cowie doesnít deny the compositionality of concepts, I wonít bother to argue for CC except to remark that, as far as anybody knows, explaining the productivity and systematicity of conceptual repertoires depends on it. That makes CC not negotiable.
        We arrive at Cowieís second objection to the Impossibility Argument.
        I hold that concepts canít be prototypes, hence that `mostí concepts must be unstructured, hence that most concepts must be unlearned. My argument for the crucial first step is that prototypes donít satisfy CC. Patently (to cite some of the classic examples) the prototypical pet fish is neither a prototypical pet nor a prototypical fish; the prototypical male nurse may be a prototypical male but is not a prototypical nurse. And so on through productively many cases. (For much more discussion, and for the argument that if concepts are prototypes, the set of counterexamples to CC is productive, see Fodor
1998a, Ch. 4.)
        To this line of thought, Cowie offers the following reply. Perhaps the possession conditions for `mostí primitive concepts have
two parts: they require both a prototype and something that determines the conceptís extension 38. Since the prototypes of a concept is, ipso facto, the intentional object of some of that conceptís ownerís propositional attitudes, learning its prototype can be a possession condition for having even a primitive concept. However, by assumption, prototypes do not satisfy CC; accordingly complex concepts need not have prototypes (and if a complex concept does, its prototype neednít be inherited from its constituents.) The situation is otherwise for the extension-determiners. Externalism being assumed, the extension of a concept is determined by causal (world-to- mind) relations; so concept possession requires the appropriate such relations to be in place. But nobody need know what they are, or even that they are, in order to have the concept in question. This is Cowieís version of the familiar externalist maxim that meanings `ainít in the head.í (The general picture derives, of course, from Putnam (1975).) Cowie expresses her two-factor account of the individuation of primitive concepts by saying that there are two senses of `meaningí (circa p. 145): meanings "in the technical sense" are required to compose. Meanings "in the intuitive sense" are what prototype theory gives an account of, and CC doesnít apply to them.
        So everythingís fine; for `mostí concepts, including most primitive concepts, the possession conditions can after all include some learning that P. IAs to the contrary not withstanding.
        This is a way out of impossibility arguments that a lot of people have suggested (including Stephen Schiffer, Christopher Peacocke, Jessie Prinz, Eric Margolis and Stephen Laurence among others). The key idea is that, although nothing belongs to the individuation of a complex concept except what it inherits from its constituents (per CC
), it doesnít follow that nothing belongs to the individuation of a constituent concept except what it contribute to its hosts. This means that there can be possession conditions for a constituent concept that are not ipso facto among the possession conditions of its hosts., and knowing the prototype for the concept might be one of these, So it could turn out that, although you canít have DOG unless you know the DOG prototype, nevertheless, you can have the DOG FROM NEBRASKA even though there is no prototype that corresponds to it; all thatís required, for the latter, is that the corresponding mental representation have the right semantic value.
        But this two factor story wonít do. For extended discussion, see Fodor
1998b, Chs. 3.4; but hereís the gist: If the possession conditions on a constituent concept C are not inherited by its host H, then it should be perfectly possible to have the latter without having the former. So (eg.) it should be perfectly possible to have the concept DOG FROM NEBRASKA without having either the concept DOG or the concept NEBRASKA 39. (A fortiori, it should be possible that someone is able to think DOG FROM NEBRASKA but not able to think either DOG or NEBRASKA and so does not find `compellingí either the inference that dogs from Nebraska are dogs, or that they are from Nebraska, or that if something is a dog and from Nebraska, then itís a dog from Nebraska.) I take that to be about as decisive as reductios get in this part of the woods.
        The upshot is that CC needs to hold in a
biconditional form: P is a possession condition on a constituent concept iff it is a possession condition on that conceptís hosts; nothing belongs to the content of a primitive concept except what it transmits to its hosts. If this is right it is very important quite aside from the innateness issues. Practically all the standard theories of conceptual content (mutatis mutandis lexical meaning, assuming that the meaning of a word is the concept it expresses) fail this strong version of CC. In particular, all theories fail according to which epistemic capacities are among the conditions on concept possession. Probably that leaves only theories that identify conceptual contents (/lexical meanings) with semantic values. These are, however, Very Deep Matters, best discussed elsewhere. (See, once again, Fodor 1998b (Chs. 4 and 5). I do wish you would read those papers. Perhaps if I were to offer a small reward? )
        Where does this leave us? Well, if CC holds in its strong form, then the Impossibility Argument is presumably ok; anyhow, itís ok for all that Cowie has to say against it. If the Impossibility Argument is ok, then `mostí such concepts as CAT. DOORKNOB and CARBURATOR arenít acquired by a learning process, which is what Iíve been trying to tell you all along.
        But now thereís really is a quandary since, if most concepts arenít learned by hypothesis formation and confirmation, why is it that so many concepts are acquired from experiences of things that fall under them? Why is it, for example, that DOORKNOB is typically acquired from experiences of doorknobs (and not, say, from experiences with cats, carburetors or pet fish?) If , as IA appears to require, the processes underlying concept acquisition are more like triggering than they are like induction,
almost anything might turn out to be the trigger for DOORKNOB.
        That is what
Concepts called the `doorknob/DOORKNOBí (=d/D) problem. The main theme of Concepts is that you have a d/D problem as soon as you accept the Impossibility Argument: for, whereas IA says that concept acquisition canít be a kind of induction, the fact that concepts are typically learned from their instance suggests that it has to be. The Impossibility Argument wants concept nativism, and the d/D problem wants concept empiricism. You canít have both so somethingís gotta give.
        Concepts offered a way of splitting this difference; one I rather like (though. of course, Cowie doesnít.) A word on this and then we really are finished. Promise.

3.5 The `Constitutioní Thesis.

        Concepts suggests an alternative to inductivist solutions of the d/D problem. True, we generally acquire DOORKNOB from doorknobs (indeed, from good (roughly, paradigm) instances of doorknobs). So be it. But maybe thatís not after all because concept acquisition is hypothesis confirmation; maybe itís because of what property being a doorknob is. The idea is that being a doorknob is mind-dependent. To be a doorknob is to have that property that minds like ours `lockí to 40 in consequence of the kinds of experiences from which our kinds of mind learn the doorknob prototype. In effect, the proposal is to do for (or to) being a doorknob what Locke did for being red (and what Humeís `second definitioní proposes to do for being a cause (Treatise, Bk. 1 Sect. XIV)); namely make it a property thatís defined relative to us. If one takes this line, then `how come DOORKNOB is generally learned from doorknobs?í is to be answered in the same way that Locke dealt with `How come itís typically red-sensations that red things cause us to have?í The answer, in both cases, is `thatís of the essence of the properties concernedí.
        As I say, this strikes me as rather a good idea; I intend, in fact, to spend a couple of more years having it. If it works, then empiricists and rationalists are both partly right about where concepts come from. The acquisition of DOORKNOB, for example, has two phases: One of them maps from (eg.) doorknob experiences to (something like) a doorknob prototype. Since prototype formation is generally held to be a species of statistical inference, this phase of concept acquisition approximates to being a rational process, just as empiricists would like. But, as weíve seen, prototypes donít compose, so they arenít the right kind of mental representations to be concepts; or even to be
components of concepts, given that CC holds in the strong form. So there has to be another stage of DOORKNOB acquisition; one that starts from a doorknob prototype 41 and yields a mental representation that is of the right kind to be the concept DOORKNOB; namely a mental representation that is `lockedí (see above) to an extension that includes all and only the doorknobs. I suppose itís just a brute fact about minds like ours that experiences of the sort that eventuate in doorknob-prototype-formation also eventuate in locking to doorknobhood; if we had different kinds of minds, weíd (as one used to say) `generalizeí differently from our experiences of prototypical doorknobs.42 Likewise, if we had different kinds of eyes, we wouldnít generalize from experiences of tomatoes to a mental representation thatís locked to being red. So said Locke, and so say I.
of course there are lots of problems with this picture; and of course the odds are that nothing of the kind will work; those are overwhelmingly the odds on any theory of mind thatís been thought of so far. But Iím unmoved by what Cowie has against it. So deeply unmoved, in fact, that Iíll take only a moment in going through her objections.
        The first is that Cowie doesnít like Lockean essences; she doesnít like properties being individuated in terms of the effects things that things that instance them have on us. She says she suspects that this kind of metaphysics must always turn out circular. (Cf. the traditional worry about Lockeís story about
being red: that it presupposes the notion of a red-sensation). But, as Cowie herself remarks, the discussion of this point has now had a couple of hundred years of being inconclusive. Perhaps it will eventually come out my (and Lockeís) way after all. Since Cowie admits to having no argument to the contrary that amounts to more than voicing a suspicion, how about if we all agree to give me the benefit of this doubt?
        Cowieís second objection begs the main issue. She says that my story about the constitution of doorknobhood and the like doesnít really give us what we want. What Cowie says we want is a psychology of concept acquisition; in particular, a theory of the mechanism whose operation explains it. Whereas, Cowie complains, I havenít provided anything like such a theory; only a (dubious) metaphysics for
being a doorknob. It is a "serious mistake" to confuse a piece of metaphysics (dubious or otherwise) with a theory of cognition. To offer the one where the other is wanted would be a typical example of philosophical a priorism.
        Indeed it would; but in fact I donít. The `constitutioní story isnít supposed to be a theory of concept acquisition; itís supposed to be an answer to the d/D problem. The
whole point of the strategy in Concepts to argue that d/D for distinguishing the d/D problem from the concept acquisition problem. According to Concepts, d/D is a metaphysical problem thatís been misidentified as psychological. What really is psychological (according to me) is not d/D but concept acquisition. Nobody knows how concept acquisition works, and Iím not expecting that anybody will find out in the next couple of weeks. But at lest we can avoid a paradox that had seemed to threaten: on the one hand, d/D gives us good reason to believe that something inductive (like prototype formation) is part of concept acquisition; and, on the other hand, the Impossibility Argument shows that concept acquisition canít be inductive. This looks like a dilemma, hence a serious embarrassment for anybody who runs a concept-based theory of mind, whichever side of the rationalist/empiricist dispute he favors. It seems, in fact, to show that thereís something wrong with RTM per se.
        But, thank goodness, the constitution story shows that it doesnít. So, like the man in Kierkegaard, weíre alright so for.
        Now, really: Did that sound to you like a moan? Or a cry for help? According to Cowie, "Fodorís position is of a kind with the mystery-mongering of Descartes and LeibnizÖ Fodor makes it admirably explicit that his `bottom lineí Ö is that acquiring concepts is a psychologically inexplicable processÖ none of the psychologistís business. (106-107)."
        Actually, if I can have Leibniz and/or Descartes for company, Iím quite prepared to monger mysteries till the cows come home. Still, the present objection is another case of Cowieís failing to grasp the polemical position. If concept formation includes a brute causal process (like the triggering of a concept by a prototype) then to that extent it is none of the (intentional) psychologistís business. But it doesnít follow that itís a mystery, or that itís `inexplicableí tout court. (While weíre at it, it also doesnít follow that it isnít.) What follows is just that concept acquisition is not a phenomenon in the domain of (intentional) psychology. Contrary, to be sure, to what intentional psychologists have generally supposed.
44 Maybe concept acquisition is a phenomenon in the domain of neurology; or physiology; or, for all I know (and for all Cowie does), geology. Any of those would surely be compatible with `the scientific world viewí. Most, indeed overwhelmingly most, things that happen in the world arenít phenomena in the domain of intentional psychology. Whatís so interesting about the mind, as cognitive science has come to understand it, is that it appears to be atypical; some of the things that happen in it apparently are. The research issue (not to be answered a priori) is which ones?
        `All right, all right; so maybe your constitution story isnít a cry for help. But isnít it still Radically Nativist? Are you a rationalist or arenít you? Damn it, why donít you `fess up?í "Regardless of what Fodor wants to call himself, the question still arises: Is he a nativistÖ.(Cowie, p. 106)" Actually, what Iím trying for is something in the middle: Empiricism is right about the relation between oneís experiences and the prototypes that having them lead one to construct. Nativism is right about the relation between the prototypes that oneís experiences lead one to construct and the concepts that constructing the prototypes trigger. Does that make Fodor Still A Radical Nativist After All? If, you positively insist that I come out of the closet hereís my very last word:
        Science is hard, theory is long, and life is short. Still, we should all do our best not to think in headlines.


1_ All Cowie references are to (1999)

2_ This is the very same Fiona Cowie who accuses rationalists in general (and me in particular; see p. 106 and passim) of having at best a "mystery mongering" account of learning on offer. Let me see if Iíve got this right: When I say that learning is a mystery, thatís me merely mongering. When she says that learning is mysterious and miraculous, thatís Cowie bravely facing up to the facts.
I wish to request a recount.

3_ This is one of the places where Cowie appears to forget that the empiricist and rationalist are equally in want of independent construals of their key notions `learnedí and `innate.í Compare Part 1.

4_ I wonít discuss Cowieís treatment of the historical figures, though I do find some of it pretty peculiar. For a quick example: Cowie thinks that Leibniz thought that you canít be an Empiricist unless you believe in metaphysically real causation. For, if you donít, "what this means, metaphysically speaking, is that [the] bearing that our experience appears to have on our mental life is strictly an illusion.(60)" But if not believing in metaphysically real causation makes you not an empiricist, then I suppose even Hume doesnít qualify. Just this once in what has been in many ways a life of self-denial, I am prepared to invoke a paradigm case argument.

5_ It is , however, not always Chomskyís way to make life easy for his exegetes. His frequent references to an innate `language organí do indeed invite the reading that POSAs are about what mechanisms are available in the `initial stateí of the language acquisition process. In fact, for reasons about to be offered, I doubt very much that that could be the intended force of the metaphor. Rather, Chomsky has it in mind to emphasize the continuity of his nativism with standard biological methodology and theory. About that he is, of course, absolutely right.

6_ In accordance with the usual practice, Iíll sometimes speak of grammars (and of UGs) as true or false, thus equivocating between grammar qua the speaker/hearerís (putative) internal representation of his language and grammar qua the linguistís theory of the speaker/hearerís (putative) internal representation of his language. Itís only the latter about which questions of truth value straightforwardly arise; but fudging the distinction helps a lot with the exegesis, and nothing essential will turn on doing so, as far as I can tell.

7_ Let a grammar of L be `descriptively adequate í iff it specifies all and only the sentences of L together with their correct structural descriptions.

8_ Iím told from time to time that the thesis that DOORKNOB is innate is prima facie very implausible. Often, the earnest tone in which this observation is proffered suggests itís a point that Iíve been supposed not to have noticed. Actually, I do understand that it seems implausible that DOORKNOB is innate. The trouble is, I find it very hard to see whatís wrong with the arguments that appear to require that conclusion. Nor do the plausibility intuitions with which several centuries of uncritical empiricism (to say nothing of a century and a half of Pop-Darwinism) have left us strike me as likely to bear much weight in the long run.

9_ This assumes what all parties to the present discussion agree about: That PAs have concepts as their constituents, and that the constituent structure of a PA is among its essential properties. Only connectionists deny this; and they wouldnít either if they could figure out some way to stop their connectionism from entailing it. (As, in fact, theyíve occasionally tried to do, but with no success. see Smolensky (1988); Fodor and McLaughlin; and Fodor (both in Fodor (1998b) ).

10_ If you are inclined to deny that there could, I suppose thatís not on account of your views about nativism/empiricism per se, but rather because you hold some form of` `theory/theoryí (or `inferential roleí theory) about the nature of conceptual content. That kind of metaphysics does entail that no concepts can be innate unless some PAs are. For present purposes, youíre welcome to whatever metaphysical assumptions about content you like. Suffice it that, unless you make some, thereís no inference from nativism about concepts to nativism about PAs.

11_ Echt laws of association are supposed to be sensitive only to spatio-temporal relations (`frequency and contiguityí) among the Ideas that they apply to. However, so hopeless is that sort of view as a theory either of learning or of thought, that empiricists have often let `similarityí and the like determine associations too. That was cheating, of course, unless thereís a domain neutral notion of similarity, (which, of course, there isnít.) Unsurprisingly, the impulse to cheat this way came back when associationism did. See (eg) the exchange between Churchland (1998), and Fodor and Lepore (1999).

12_ One might argue that the kind of knowledge that explains linguistic capacities is `knowing howí not `knowing thatí, hence that having it doesnít require believing or cognizing anything. But such a view leads to the rejection, not just of a mentalistic reading of nativism, but to a mentalistic reading of empiricism as well. It is therefore not Cowieís line. Cowie wants rationalism to be false compatible with the cognitive turn in psychology having been Quite A Good Thing.

13_ Note how Plato (himself a bit of a Platonist) explains the slave-boyís ability to do geometry in the Meno. Holding that what one knows explains oneís capacities is entirely compatible with holding that the objects of oneís knowledge are non-natural.

14_ A puzzling passage this. One might have thought that I just couldnít have a better reason for preferring my theory to yours than that yours doesnít exist. (Assuming, of course, that mine does.) Notice, by the way, how much the "yet" is tendentious. Likewise the "real" in the sentence that follows .Cowie is rather prone to obiter dicta about what "really" explains what.; see below.

15_ My wife is in this line of work, and she assures me that is so. Maybe Cowie should go argue with her.

16_ More precisely, in the PLDs from which they could acquire language, consonant with the normality of the process. Critical experiments, in which the conditions of language acquisition are systematically controlled, are of course not possible; so the distinction between what is merely typical of the acquisition process and what it actually requires is hard to draw. This is a kind of point of which ethologists are forever reminding us: Birds "learn" to fly if they are given normal opportunities to practice. But, as it turns out, they also "learn" to fly if theyíre not.
That Cowie is inattentive to this caveat is hardly surprising. If you think of languages the way she prefers to, viz not as things people know but as "spatiotemporally located natural objects" youíre correspondingly unlikely to think of linguistics as responsible to the counterfactuals about what human languages there could be, or the conditions under which humans could acquire them. Cowie says literally nothing about whether she takes linguistics to be responsible to such counterfactuals, or about what she thinks their truth-makers are.

17_ A great lot of the cross-disciplinary discussion of Chomskyís theory has turned on whether the PLD reliably exhibits sentences whose derivations require structure-dependent operations. Thatís what Chomsky gets for offering an example thatís easy to understand. It therefore bears emphasis that structure dependence is only one of very many constraints that UG is supposed to impose upon grammars; hence to which the PLD must testify if the thesis that children approach the PLD with a UG already in mind is to be supposed untrue.

18_ We Ancients remember `subtle cuesí very well. They used to pop up whenever, on the one hand, a psychologist was hell-bent to explain the organization of a creatureís behavior by appeal to the structure of its environment; but, on the other hand, a survey of the creatureís environment failed to reveal psychophysical counterparts of the structure it was presumed to have. Thus Skinnerian behaviorists thought there must be some `stimulus propertiesí that are reliable indicators of (as it might be) the bankruptcy of a financial institution; because, after all, some human organisms (viz. accountants) are able to respond in a way that discriminates bankrupt institutions from others in environments that contain the relevant account books. Just what `stimulus propertyí controls such selective responses as `the capitalization would appear to be inadequately fluidí remains, to be sure, a matter for further research. Beyond doubt, itís one of those `subtle cues.í
So nice to have them back.

19_ There is no indication, either in Cowie or in the literature she cites, how such information (about transition probabilities among phonemes, as it happens) might be employed to isolate anything thatís grammatically pertinent except morpheme boundaries. The rest of a statistical theory of language learning has "yet" to be "worked out."

20_ Rather oddly, Cowie appears to hold both that thereís no case for the childís lack of negative information in language acquisition and that "there is a dearth of negative evidence in every domain in which people learn. .For example, you learn what Curry is without being told about all the things that curry isnít (215, my italics)." In fact, Cowie remarks, "human beings learn an awful lot, about bewildering variety of topics, from sketchy and largely positive data. That they can do soÖ is miraculous and mysterious. It is not, however, a reason to accept a nativist explanation of the miracle as the solution of the mystery" (216). She doesnít, however, say why itís not except for remarking that "itís just absurd to suppose that the domain-specific principles required for learning about curries are innate (215). "Why, one wonders, does Cowie think so? It looks like whatís absurd isnít supposing that learning about curry requires lots of information that is innate and domain specific, but rather supposing that curry is the domain to which the innate information is specific. (Try food; and see the introduction of practically any serious cookbook; where thereís likely to be an attempt to make some of the relevant domain-specific generalizations explicit.) Likewise, nativists about language donít suppose that the domain of the innate information thatís used to learn English is English; they only claim that English is in that domain. Itís an open, empirical question --- indeed, one that linguistics is devoted, almost entirely, to answering--- what else is in there too (what it is that all and only the possible natural languages have in common.)
Reductio only works on arguments with false conclusions.

21_ Empiricists are forever giving nativists edifying lectures on this point. Thus Cowie: "Conservative politicians, moralists, and jurists apparently find overwhelming the inference from `innate` to `rightí and `inevitableíÖ [But] to suppose that something is right just because it is innate is to commit the fallacy of deriving `oughtí from `isí Ö. the inference is Ö being made all the time, with potentially devastating consequences. (x-xi; see also Ellman et al))."
I do find this sort of special pleading extremely offensive.

22_ Bear in mind, by the way, thatís not all a child has to keep him busy. He has to learn a lot of vocabulary too. To say nothing of the geometrical structure of perceptual space, the intuitive physics of middle-sized objects; the intuitive intentional psychology of his conspecifics, and so on; all of which information enlightened empiricists, just like their unilluminated empiricist colleagues, presumably take not to be genotypically carried. (Also, infants sleep a lot.) Since Cowie sticks exclusively to the issue of nativism about language, she is never required to discuss the overall plausibility of the empiricist view of cognitive development. It is probably wise of her not to do so.

23_ Notice, in passing, that the PLD is better evidence for the grammar of L than it is for UG; which suggests (pace EE) that if both are learned, itís the grammar, and not UG, that should be learned first. Lís grammar expresses the structural similarities that sentences exhibit in virtue of their all belonging to L. UG, by contrast, expresses only the very abstract structural similarities that sentences exhibit in virtue of their all belonging to some natural language or other. (Read `some languageí with short scope relative to the `allí.)
Parallel considerations very strongly suggest that the grammar-acquisition mechanisms should be domain specific even if theyíre not innate: The chances are surely overwhelming that arbitrary sentences drawn from a PLD will be more similar to one another (hence better evidence for both the grammar and for the UG) than they are to arbitrary objects of the childís nonlinguistic experiences. Any sentence of English is ipso facto more similar to another English sentence and to any sentence of Russian then either is to a bird, or Mother, or a jar of emulsified peaches. (That, I suppose, is what it is to think of language as a domain.)Yet it is regularities among a childís nonlinguistic experiences that his inductions from data to the UG depend on according to EE. "If you assert, therefore, that the understanding of the child is led to this conclusion by any process of argument or ratiocination, I may justly require you to produce that argument, nor have you any pretense to refuse so equitable a demand". (Hume, INQUIRY, Section IV, Part 1).

24_ Since P entails P, itís not clear that EE adds anything at all to the assumption that the child canít learn the language unless he knows P.

25_ Cowie appears to hold that iteration arguments somehow require as a premise that UG is psychologically real (see p. 273). But she is quite wrong to think that; as, indeed, the present discussion shows. Iíve used nothing about the ontological status of UG; Iíve assumed only what EE grants: that you canít learn a first language unless you know that grammatical rules are structure dependent.

26_ It testifies to Kantís genius that he saw that a cognitive theory that posits across-the-board structural preferences, is in need of a transcendental argument that the world can be relied on to comply with them. But most people doubt that he actually had one.

27_ Itís a considerable irony that the notion of having a concept that Chomsky thinks is needed for such very special purposes as explaining language learning, is much the same one that philosophers who confuse metaphysics and semantics with epistemology think is satisfied by the possession of empirical concepts quite generally: They think that to have a concept is to know `in principleí how to identify the things it applies to, and/or to know `in principleí how to bring about states of affairs in which the concept applies. My guess is that Cowie has taken some such verificationist view of concept possession for granted. Putting that together with the observation that, in the general case, concept acquisition doesnít require a grasp of essences, gives Cowie the argument presently under examination.

28_ Typical avatars of this venerable confusion include the various semantic holisms that so many empiricists now endorse. (The latest being the `theory theoryí of concept individuation, according to which the identity of a concept is determined by the beliefs it is embedded in rather than the other way Ďround.) In semantics (as elsewhere), outbreaks of holism are invariably signs of a foundational blunder. For discussion of this complex of issues, see Fodor and Lepore (1992).

29_ Weíve already seen one reason why 3.2.4 must be true; viz that concepts are ontologically prior to the kinds of things that can be (dis)confirmed. The argument now unfolding waives that objection and assumes, for the sake of the discussion, that the notion of concept learning by hypothesis testing is coherent; but it claims, even so, that primitive concepts canít be learned that way.

30_ Cowie apparently thinks that "deferential" concepts are somehow an exception. My own view is that the category belongs to sociology, not semantics (see Fodor (1994)). But it will do for present purposes that you canít reason with a concept unless somebody already has it. Thatís a truism, no?

31_ One is reminded of a familiar parody of lawyerly arguments: `My client didnít do it, he wasnít there; and even if was there, he didnít have a gun; and even if he was there and had a gun, it wasnít loaded; and even if he was there and had a gun and it was loadedÖ. usw.

32_ Which is not to deny what `methodological solipsismí claims: viz. that a mental processes applies to the concepts in its domains "solely in virtue of" their nonsemantic properties. That mental processes are syntactically driven doesnít at all imply that metal representations are syntactically individuated. I canít begin to tell you how many philosophers have been confused about this over the years.

33_ As Iíve already suggested, this line of thought depends on taking for granted the (fashionable) externalist view of the supervenience base for semantic properties. Letís grant this assumption, for which, however, Cowie offers only the following strange argument: "Ö protoconcepts [canít] be conceptual roles understood dispositionally as networks of potential [sic] causal/inferential interactions. For protoconcepts are supposed to be innately specified, whereas the dispositions that our concepts have to interact causally or inferentiallyÖ are not. I was not born such that my tokenings of PLATYPUS are disposed to cause tokenings of MONOTREMEÖ. (85)" This is, I think, the only argument Iíve ever heard against a conceptual role semantics that doesnít work. The most it shows is that somebody who is both an internalist and a nativist about concepts shouldnít also be an unmitigated semantic holist. Rather, he ought to hold that a conceptís innateness requires only the innateness of its constitutive inferences; and, by assumption, much less than every inference that a concept enters into is constitutive. In fact, practically every internalist does hold something like this (often at the price of endorsing an analytic/synthetic distinction.) So, whatís the problem about internalists being nativists?
Notice, in particular, that not even every necessary inference that involves concept C is is ipso facto constitutive of C (assuming constitutive inferences to be the ones that correspond to possession conditions). So, even if you are an inferential role internalist, and even if you think that PLATYPUS is innate, and even if you think that a platypus is a monotreme is necessarily true, you still are not required to claim that a platypus is a monotreme is innate; nor are you required to claim that having the concept MONOTREME is a possession condition for having the concept PLATYPUS; a fortiori, you are not required to claim that having very concept that interacts with PLATYPUS interacts causally is a possession condition for PLATYPUS. Concepts cautions against this mistake, oh, maybe fifty times. To no avail, it appears.

34_Iím not, of course, supposing that anything so simple would work as a metaphysics of the content of innate concepts; just that the proposal is perfectly bona fide qua externalist. If you want an externalist metaphysics of the content of innate concepts thatís not just bona fide but true, Iím afraid there isnít one "yet". (But, there isnít one for learned concepts either; or for words. I donít suppose thatís an argument that there arenít any words or concepts.)

35_In passing, and quite independent of issues about nativism: thereís every reason for an externalist to take content to supervene on possible (including nonactual) causal relations. He thereby disencumbers himself of such embarrassments as Donald Davidsonís `Swampman.í (According to Davidson, since Swampman has no causal history, he ipso facto has no intentional states.) In semantics as elsewhere, whatís actual doesnít matter much to the metaphysics; itís the counterfactuals that count.

36_ I assume (as does Cowie) that the other familiar account according to which `mostí concepts are complex ---viz that they are definitions--- is no longer seriously in the running. For arguments, See Fodor (1998a, Ch.3).

37_ For simplicity, I ignore such content properties of C1 as may be determined by the arrangement of its constituents (i.e. by its `syntaxí). This is, to be sure, no small matter; itís presumably such arrangement features that distinguish (eg.) A PERSONíS FAVORITE CAT from A CATíS FAVORITE PERSON. But abstracting from the effects of syntax on conceptual content wonít affect our present purposes, and it simplifies the exposition.

38_ Thereís some question whether the `secondí component of a concept is to be an extension determiner, or an extension. My own view (but not Cowieís, as far as I can make out) is that extensions are much the better if those are the choices. For one thing, itís a lot more plausible that extensions compose than that the world-to-mind relations do that are supposed by externalists to be what fix semantic values. For another thing, it would be nice if the content of a concept were ipso facto at least part of what the concept expresses. This will be so if contents are semantic values, but not if they are the mechanisms that mediate world-to-mind connections. Maybe DOG expresses the set of dogs, or the property of being a dog, or the like. But it certainly doesnít express whatever the causal hookup is that, by assumption, connects DOG to dogs or to dogness.

39_ To say nothing of not having the concept FROM. Prototype theorists tend to keep an extremely low profile in respect to the possession conditions for concepts that express relations; as well they might since, prima facie anyhow, relation concepts would seem not to have prototypes. Cowie doesnít discuss the issue.

40_ `Lockingí is a place holder for your favorite (externalist) theory of the relation such that, if it holds between a thought type and a property, then the property is the intentional object of tokens of that thought type.

41_ However, see the footnote before last. God only knows what, if anything, corresponds to prototypes in the case of relational concepts.

42_ Reading `prototypical doorknobsí rigidly; i.e. as prototypical of actual-world doorknobs.

43_ He has fallen from a window and is half way down.

44_ More precisely, contrary to what many of them have supposed themselves to suppose. If you look closely at paradigm empiricist/constructivist theories of concept acquisition, itís not at all clear to what extent they endorse intentionalist solutions. Locke and Hume thought that you get RED `brute causallyí from stimulations of the sensorium. Dewey thought you get new concepts by doing some "doing and undergoing"; (whatever that is, it doesnít sound much like anything intentional.) Piaget thought you get them by doing some "assimilating and accommodating.," (to which the same applies.)


Block, N. (1980) Readings in The Philosophy of Psychology, 2 vols. Harvard University Press, Cambridge, MA.

Chomsky, N. (1965) Aspects of The Theory of Syntax, MIT Press, Cambridge, MA

Chomsky, N. (1972) Language And Mind, Harcourt Brace Javonovich NY.

Churchland, P.M. (1998) "Conceptual similarity across sensory and neural diversity," in Churchland, P.M. and Churchland, P. On The Contrary, MIT Press, Cambridge, MA.

Cowie, F. (1999) What's Within? Nativism Reconsidered, Oxford University Press, NY.

Elman, J. et al (1996), Rethinking Innateness, A Connectionist Perspective on Development, MIT Press, Cambridge, MA.

Fodor, J. A. (1981) "The Present Status of the Innateness Controversy", in his Representations, Cambridge, MA: MIT Press.

Fodor, J. A. (1994) The Elm and The Expert, MIT Press, Cambridge, MA

Fodor, J. A. (1998a) Concepts: Where Cognitive Science Went Wrong. New York: Oxford University Press.

Fodor. J. A., (1998b), In Critical Condition, MIT Press, Cambridge, MA.

Fodor. J. A., (1998b_4), "There Are No Recognitional Concepts - Not Even RED" Chapter 4 in (1998b)

Fodor. J. A., (1998b_5), "There Are No Recognitional Concepts - Not Even RED, Part 2: The Plot Thickens" Chapter 5 in (1998b)

Fodor, J. A., (1998b_10) "Connectionism and the Problem of Systematicity (continued): Why Smolensky's Solution Still Doesn't Work", Cognition 1996, 62, 109-119, reprinted as Chapter 10 of (1998b).

Fodor, J. A. (1998c) Unambiguous triggers,í Linguistic Inquiry, 29.1, 1-36.

Fodor, J. A. (2000) The Mind Doesnít Work That Way, MIT Press, Cambridge MA.

Fodor, J. and Lepore, E. (1992), Holism, A Shopperís Guide, Oxford, Blackwell.

Fodor, J. and Lepore, E (1999) "All at sea in semantic space; Paul Churchland on meaning similarity" Journal of Philosophy XCVI, no. 8, August. 381-403.

Fodor, J. A. and McLaughlin, B. (1990) "Connectionism and the Problem of Systematicity: Why Smolensky's Solution Doesn't Work", Cognition, 35, 183-204, reprinted as Chapter 9 of (1998b).

Karmiloff-Smith, A. (1992) Beyond Modularity, MIT Press, Cambridge MA.

Nozick, R. (1981) Philosophical Explanations, Harvard University Press, Cambridge, MA.

Putnam, H. (1975), "The meaning of 'meaningí", in Gunderson, K. (ed.) Minnesota Studies in The Philosophy of Science, University of Minnesota Press, Minneapolis, Minn.

Smolensky, P. (1988) "The proper treatment of connectionism,: Behavioral and Brain Sciences, 11, 1-23.

 top |PARTS 1-2| back to symposium index | Cowie's reply