CONCEPTS: WHERE COGNITIVE SCIENCE WENT WRONG. By Jerry A. Fodor. New York: Oxford University Press, 1998. Pp. xii, 174.

Kent Bach

As the dust jacket proclaims, “this is surely Fodor’s most irritating book in years. ... It should exasperate philosophers, linguists, cognitive psychologists, and cognitive neuroscientists alike.” Yes, Fodor is an equal-opportunity annoyer. He sees no job for conceptual analysts, no hope for lexical semanticists, and no need for prototype theorists. When it comes to shedding light on concepts, these luminaries have delivered nothing but moonshine. Fodor aims to remedy things, and not just with snake oil. He serves up plenty of clever barbs, potshots, and one-liners, not to mention arguments, to promote his “informational atomism.” It states that there is a large class of concepts, namely lexical concepts, which are ontologically and semantically primitive: each such concept has no internal structure and has its content not in virtue of any relation it bears to other concepts but in virtue of a nomic relation it bears to some property. The property it is thus “locked to” is its content.

The first chapter sets the background by sketching Fodor’s representational/computational theory of the mind (RTM), according to which thinking is computation, the contents of mental states explain their computational roles, and mental representations are the primitive bearers of content (7-10). Moreover, “meaning [content] is information (more or less)”: a concept expresses a certain property in virtue of its being a law that things with that property cause tokenings of that concept in certain “still-to-be specified circumstances” (12). When different concepts express the same property, what distinguishes them is “in the head” (15), the different “modes of presentation” associated with them. But whereas for Frege these are senses, for Fodor they are syntactic, neural analogs of shapes. The main thrust of the book is that informational atomism is the only cogent theory of concepts that comports with RTM.

In the second chapter Fodor makes five “non-negotiable” assumptions about concepts (23-28): (1) concepts are mental particulars (and so can function as mental causes and effects), (2) things “fall under them,” (3) they are constituents of thoughts, whose contents are determined by the contents of their constituents and how these constituents are combined, (4) some—but only some—concepts must be learned, and (5) concepts are public and shareable. (1) is puzzling: if one’s cat concept is a particular, how can one have different thoughts about cats all containing that very particular? And if one thinks, say, that female cats are tamer than male cats (or one has two concurrent thoughts about cats), then somehow that particular has to occur in two places at once. Fortunately, the distinction between concept tokens and concept types, which Fodor notes (3n.) but tends to ignore, is helpful here. Whereas assumption (1) concerns concept tokens, (2), (4), and (5) obviously apply only to concept types (assumption (3) can be taken either way, since there are both thought tokens and thought types). Similarly, he must be speaking of concept types when he says that concepts “have their contents essentially” (120).

Since any particular is a token of innumerable types, the question arises of how concept tokens are to be type-identified, presumably for the purposes of psychological explanation. Having the same content (reference) is, as we learned from Frege, necessary but not sufficient. Unfortunately, Fodor’s notation makes it seem as if this is sufficient. He refers to concepts by using locutions like ‘the concept COW’ and ‘the concept RED SQUARE,’ as if putting ordinary words or phrases in capital letters uniquely determines a concept (putting them in boldface italics won’t do the trick either). He notes that “the expressions appearing in caps are names, rather than structural descriptions” (41n.), but that doesn’t guarantee uniqueness, much less identify which concept is being named. If the choice of concept name is based on the meaning of the ordinary word, then which concept does, say, ‘the concept HARD’ (or MODEL, PLANE, or PITCH) refer to? That is a hard question.

The next three chapters are largely polemical. They reinforce Fodor’s previous arguments against prototypes, definitions, and constitutive inferential roles. He begins, though, by denouncing lexical semantics and ridiculing its notions of polysemy and lexical structure (and decomposition). For example, he debunks an account of the alleged lexical structure of the tricky verb ‘keep’ (‘get,’ ‘make,’ and ‘put’ are similarly tricky) as it occurs in such diverse phrases as ‘keep your receipt,’ ‘keep your clothes on,’ and ‘keep washing your hands.’ Fodor offers an atomistic alternative: ‘keep’ expresses, well, the concept KEEP. Here one can only marvel at how human psychology manages to lock to the unique property this concept picks out, the property of keeping. Presumably this is a distinct concept from the concept RETAIN, which must pick out a different property in RETAIN YOUR RECEIPT, since you can’t retain your clothes on or retain washing your hands (and there’s a big difference between retaining a lawyer and keeping a lawyer). Also, when debunking the notion of polysemy Fodor seems to assume that it’s straightforward how concepts compose. In fact, there is a legitimate problem, analogous to one that lexical semanticists worry about, of how to explain the different ways in which, e.g., the concept FAST contributes to the contents of FAST CAR, FAST DRIVER, FAST TRACK, FAST RACE, and FAST TIME.

It’s hard to dispute Fodor’s complaints with prototype theory, at least as a theory of concepts, especially his objection that it can’t account for the contents of composite concepts, such as his pet concept PET FISH. Prototype theory may help explain people’s conceptions (as opposed to concepts) of (typical) things of different types and various facts about how people categorize things, but it isn’t serviceable as a theory of concepts, since prototypes don’t compose and concepts do. As for definitions, Fodor claims that most lexical concepts (ones for which we have single words) don’t have them. Surely he right to think that at least “some concepts are going to have to be primitive” (130), but he might have considered the possibility that many lexical items are hard to define because they are associated, intra- or interpersonally, with several distinct but closely related concepts. And don’t bother trotting out examples of concepts that do have definitions, like BACHELOR, EFFECT, ISLAND, NAVEL, TRAPEZOID, VIXEN, and WEEK, for Fodor insists that even these are unstructured, i.e., they are not composed of the concepts that (allegedly) define them. For example, contrary to Kant’s conception of “conceptual containment,” the concept VIXEN does not have the concepts FEMALE and FOX as constituents.

Not only does Fodor deny that (most) lexical concepts have constituent concepts, he denies that they have constitutive inferential roles. Evidently he takes this to follow from Quine’s objections to the analytic/synthetic distinction, although he never explains which of Quine’s arguments, if any (surely not the behaviorist ones), both transfer from language to thought and cohere with the sort of representational theory of the mind that Fodor himself endorses. Anyway, he denies that, for example, the concept VIXEN entails the concept FOX. Of course to be a vixen is to be a fox, indeed a female one, but that’s a fact about the property of being a vixen, not about the concept VIXEN. The non-lexical concept FEMALE FOX is not atomic in either of these senses. Evidently, then, anyone whose lexicon does not contain the word “vixen” or a one-word synonym for it does not possess the concept VIXEN. Monolingual speakers of a language lacking a such word are not in a position to have that concept, although they could still have the concept FEMALE FOX. And it is perfectly possible to have the concepts VIXEN and FEMALE FOX without believing that vixens are female foxes. In short, Fodor is claiming that what are traditionally regarded as (conceptually) necessary connections between concepts are really (metaphysically) necessary connections between the properties picked out by concepts. “To be a vixen is to be a female fox” expresses a fact about vixenhood, not the concept VIXEN. Accordingly, people’s insistence that they wouldn’t “call” something a vixen if it were male or that it wouldn’t “count” as a vixen if it were male reflects their metaphysical knowledge about vixenhood, not any conceptual knowledge. There is nothing incoherent or “conceptually confused” (as some philosophers used to say) about thinking that vixens can be male. If you think that you’re making a metaphysical mistake, but at least you’re clear on the concept.

So for Fodor, “intuitions of conceptual connectedness are ... an illusion” (86). One wonders, though, if concepts have their contents essentially, then why can’t one concept entail another (in the sense that to instantiate the one is necessarily to instantiate the other)? For example, if CAT has as its content the property of being a cat and ANIMAL the property of being an animal, why can’t CAT entail ANIMAL? Of course this assumes that having the property of being a cat entails having the property of being an animal but, at least on my CAT and ANIMAL concepts, it does. For someone, like Fodor, relative to whose CAT and ANIMAL concepts having the property of being a cat does not entail having the property of being an animal, there is no such conceptual entailment. Of course, for him ‘being an cat’ expresses a different property than it does for me. But that’s my point: the essential difference between our cat concepts corresponds to the difference in the properties they refer to. Notice that the relation between my concepts CAT and ANIMAL is an external conceptual relation, not the dreaded internal conceptual relation (conceptual containment) that Fodor thinks lexical concepts don’t enter into. But it is not merely a constitutive inferential relation. It is a metaphysical relation between the property of having one concept and the property of having another, not unlike the metaphysical relation between the two properties. If concepts (types, not tokens) can have their contents essentially, why can’t they have their relations essentially? And if a concept essentially has its content because it is nomically related to its content, why can’t it be essentially related to another concept because it is nomically related to that concept? That is, why can’t one concept be such that it wouldn’t and couldn’t apply to something unless another concept also did?

In the last two chapters, Fodor explores the implications of informational atomism, whereby (most) lexical concepts are unstructured and have their contents in virtue of being nomically related to properties. This is how we can gain “semantic access” to properties (25). Now you might want to know what it is for a concept to be thus “locked to a property” (not to mention how to go about finding out that a concept is locked to one property rather than another). Fodor suggests that there are various ways in which this can happen, but all we get is a schema of a possible story. He seems more interested in sketching, in as empirically noncommittal a way as possible, what he takes to be the only alternative to views that treat concepts as structured, as essentially related to other concepts, or as having constitutive inferential roles. Whatever the true story may be, on Fodor’s view lexical concepts are not learned. According to Fodor, learning can only involve hypothesis formation (evidently feature detection, pattern recognition, and abstraction can’t result in learning), but that would require using the very concepts supposedly to be learned, since lexical concepts don’t have constituent concepts in terms of which the relevant hypotheses would be couched (124). However, concepts are not innate in the sense of having their contents independently of any encounters with the environment. Rather, interaction of the right sort “triggers” them into having their contents, by somehow locking them to certain properties. In this regard, it is unclear whether Fodor thinks that each concept is destined to have a certain property as its content—this would seem to make the triggering superfluous—or whether a concept gets recruited for its job.

Since most properties of things do not stand in nomic relations specifically to us, what sorts of properties can concepts lock to? This is the “doorknob/DOORKNOB” problem (Fodor perversely uses this is his paradigm despite noticing (122n.) that DOORKNOB “is plausibly a compound”). He suggests that the properties to which most concepts are locked are in a certain sense mind-dependent. By this he means not that they are in the head but that they are relational, individuated by how they strike us (136). Our “mental structures contrive to resonate to [e.g.] doghood” (80). Thus the locking relation (whatever it is) to such a property is instantiated because of the way we are. This is somehow true even for DOORKNOB and doorknobhood and BACHELOR and bachelorhood. Unfortunately, Fodor does not tell us how the story is supposed to go for concepts expressed by verbs, adverbs, or prepositions, which are conspicuously absent from the discussion. But he does hold that some concepts, like GENE and NEUTRINO, are locked to properties in a more indirect way. The properties they are locked to are mind-independent. These are natural kind concepts, concepts of natural kinds “as such,” and acquiring them (locking to the properties they express) requires the help of theories. It’s not clear which category WITCH and UNICORN fall into.

At any rate, most lexical concepts are like the concept RED in that they pick out “appearance properties,” ways in which things can strike us . These are not limited to sensory properties. Our minds lock (or is it Locke?) to such properties as redness, rockhood, and doorknobhood because each of these properties comprises a distinctive way in which we can be struck by things. Experiences with “stereotypic” cases somehow cause us to lock on to the property in question. At least this is a sufficient condition for locking to a property. It’s not clear how this process could work for ANIMAL, MINERAL, or VEGETABLE, never mind relational concepts. Also, there’s the question of which property a stereotypic case is a stereotype of, for it seems that different properties can have the same stereotype (Fodor says nothing to suggest otherwise). So how does a stereotype, even with the help of human psychology, determine a unique property and thereby the reference of the concept it triggers?

Details aside, it does seem that the world must be nomically related to the mind somehow if the mind is to get a grip on the world. So Fodor is surely on to something, perhaps even locked to it. But just what that is is not entirely clear.