SOFIA, Lisbon, 21-24 May, 1994
Can Semantic Properties be non-causal?
(Comments on Fodor's "Concepts - a Tutorial Essay")
As you must have gathered from Jerry's paper, he is not really enthusiastic about what he calls Pragmatist accounts of concepts. By his lights, a Pragmatist account claims that having a concept is "being able to do certain things rather than being able to think certain things" (p. OOO). This Pragmatist view, he says, is typical of Current (or Contemporary) thinking about concepts. And it is deeply wrong.
Now, it is tempting to question Jerry's contrast between thinking something and doing something, on the grounds that thinking something can't be doing nothing. On second thought, however, and on behalf of Jerry's anti-Pragmatism, it might well be that a person's thinking a thought does not reduce to anything else the person may do. Perhaps it does not reduce to any non-thinking behavior in which the person may engage. Jerry's anti-Pragmatist stance would then merely amount to an innocuous aversion to Behaviorism. Nor would the identity between a person's thinking something and the electrical activity of her brain cells vindicate a Pragmatist theory of thinking, since the electrical activity of a person's brain cells is not something the person does. Rather, it is something which occurs within her.
However, I am not interested in pursuing this line of thought as such. For I take it, what Fodor objects to is a Pragmatist theory of concepts, not a Pragmatist theory of thinking. The issue raised by Fodor, I think, is not whether thinking something is or is not doing something. The question rather is whether a system can have a concept and not do anything with it. The issue is whether a concept can stand by itself in a creature's mind without a network of surrounding concepts, without entering a system of beliefs and propositional attitudes, without being part of a number of epistemic capacities, without entering inferences or inferential processes.
Granted, the issue is not whether thinking a thought is or is not doing something or other. There is, however, a slightly different question which, I think, is relevant to Fodor's project. The question is: What is the difference between thinking about Xs and being able to reason, to make inferences and to recognize Xs? In the process of exploring this slightly different question, I would like to suggest that there is a tension in Jerry's thinking: on the one hand, Jerry subscribes to an informational view of the content (or meaning) of primitive concepts. On the other hand, in this paper, he argues for keeping the semantic properties of symbols apart from their causal properties. This is the content of his claim that semantics is not part of psychology. But since on any informational account, the content of a symbol depends on its causal relations with properties in the environment, one may wonder whether Jerry can divorce the causal and the semantic properties of concepts as he wants to.
To start with, notice a crucial difference between Jerry's notion of a concept and the Fregean tradition. In the Fregean tradition, a concept is a sense (or mode of presentation of an object) - something abstract which is expressed by a symbol (e.g., a word of a natural language). Presumably, in the Fregean tradition, concepts do not have causal properties. But, on Jerry's view, they do because concepts are symbols in the language of thought.
With Fodor, let's distinguish primitive (undefinable) concepts and complex concepts built out of the former. Not only complex concepts, but also propositional attitudes are built out of primitive concepts. Primitive concepts are the simplest kind of things with both semantic and causal properties. Now, I will first try and characterize accurately Jerry's present position. I will ignore logical and mathematical concepts or concepts without an extension (I will assume that concepts without an extension can't be primitive).
Although concepts have both semantic and causal properties, only their semantic properties, not their causal properties, are relevant to the individuation of primitive concepts. Presumably, on Fodor's present informationally-based view of concepts, a primitive concept derives its representational and misrepresentational powers from the Asymmetric Dependency Condition. Now what of coextensional concepts? On the assumption that they derive their representational and misrepresentational powers from the Asymmetric Dependency Condition, aren't they bound to share all their semantic properties? Well, on Fodor's view, there is still room for distinguishing two coextensional concepts: coextensional concepts can still be distinguished by the syntactic properties of their respective vehicles, specifically, by the constituent (compositional) structure of the distinct vehicles (the distinct mental symbol-types or vehicle-types) expressing the concepts. So the concept CREATURE WITH A HEART will differ from the concept CREATURE WITH A KIDNEY in virtue of the fact that the vehicle of the former includes the vehicle "heart" and what expresses the latter includes the vehicle "kidney". The concept WATER differs from the concept H2O in that one can have the former, not the latter, without having the concepts HYDROGEN, OXYGEN and NUMBER 2. It follows, of course, that distinct coextensional concepts whose respective vehicles would not have different constituent structures are precluded by Jerry's view.
What is distinctive of Fodor's present informationally based view of the nature of primitive concepts is that, as he likes to put it, "semantics is not part of psychology". It's the business of semantics, not psychology, to individuate concepts. So the division between semantics and psychology merely reflects the difference between the semantic and the causal properties of concepts. The task of semantics is to understand what makes it possible for elementary and complex mental representations to have the power to represent things. In virtue of what does some physical object (or some state of a physical device) have the ability to represent (and misrepresent) things and states of affairs in the world? Where do mental representations derive their power to represent things from? How could a purely physical system represent and misrepresent things out there? This is the task of a naturalistic semantics. In a sense, naturalistic semantics - for the language of thought - has been elevated to the rank of metaphysics. Fodor wants to know how symbols in the language of thought hook up onto the world. And you don't need psychology to do this. (Nor do you need epistemology.) The task of psychology is to produce a theory of the underlying causal mental processes turning a thought into another thought or leading from thought to intentional behavior. Given that, on Fodor's view, the causal properties of mental representations are syntactic or formal (i.e., non-semantic), we have a nice division of labor: psychology is computational. It deals with causal processes. Semantics has nothing to do with it. It deals with the representational properties of symbols.
Mirroring the distinction between semantics and psychology - the distinction between the semantic and the causal properties of concepts - is the distinction - central to Jerry's present paper - between concept-individuation and concept-possession. The two main theses of Jerry's present paper are the following:
1. Every theorist faces the following option: either concept-individuation is derivative from (or parasitic upon) concept-possession or vice-versa concept-possession is derivative from (or parasitic upon) concept-individuation. For cats, says Jerry, it's clear which should come first: first, you individuate cats; and then you say what having one of those consists in. For pains and jobs, he concedes, it goes the other way around: first, you say what it is to have one; and individuation follows.
2. For concepts, individuation comes first; possession comes next.
Classical theories derived concept-possession from concept-individuation. And they were right to do so. Jerry's prototype of Classical theorizing about concepts is Hume. On Hume's view, concepts derived their semantic properties from Resemblance. Causal properties of concepts depended on laws of Association. Now, Fodor thinks that Hume's account of both semantic and causal properties of concepts was wrong. On Fodor's view, semantic properties - as I said - are informational; and causal properties are computational. As he has said somewhere else: Dretske and Turing between them have solved the problems of intentionality. But Hume was right to give priority to concept-individuation over concept-possession.
On his view, Current criticisms of Classical theories have completely missed their target. What they should have done is reject Hume's particular account of the semantic and causal properties of concepts but keep the priority of concept-individuation over concept-possession. Instead, they have taken the other option: they have given priority to concept-possession over concept-individuation. And they are deeply wrong to have done so. In Jerry's words: "this subtle and largely inarticulate difference beteen Contemporary RTM and their Classical forebearers has... the most profound implications for our cognitive science... I suspect that it was a wrong turn - on balance, a catastrophe...".
What Jerry calls a Pragmatist view of concepts is a view which gives priority to concept-possession over concept-individuation. Presumably, the reason why Jerry calls such a view a Pragmatist view is that a view which gives priority to concept-possession over concept-individuation is bound to individuate a concept by means of what a creature who possesses the concept is likely to do with it. By contrast, Jerry would say that a view which gives priority to concept-individuation over concept-possession would individuate a concept by saying what it is a concept of. A Pragmatist view, then, is a view which says that semantics is part of psychology. It might - in a Fregean style - define concept-possession by the capacity to have certain beliefs and other propositional attitudes. And so a concept would then be individuated in terms of the beliefs whose complex contents contain it as a constituent. From Jerry's perspective, this is unacceptable since it falls prey to Meaning Holism, which, in turn, is inconsistent with scientific psychology.
Now, I'd like to make four points. First, I'd like to ask Jerry for clarification of a textual matter. In the course of reconstructing the misdirected criticisms of Current thinking to the Classical view, he lists three objections - methodological, metaphysical and epistemological. It's the third kind of objection which raises a puzzle in my mind. As I understand it, it goes like this: what is wrong about the Classical picture (from Descartes to Kant) is that it postulates a system of internal representations - a Veil of ideas. A representation (something mental) can only make contact with other representations (other mental things). So the question arises: How can one regain contact with the world after one has lost it by postulating the Veil of ideas? And Fodor suggests that giving priority to concept-possession over concept-individuation was then thought of as a way out of the Veil of ideas. This is what I don't get: Assuming that the problem is to get out of the bottle of mental representations, then how could giving priority to concept-possession show the fly the way out of the bottle of mental representations? Why should a system of epistemic capacities make it any easier to contact the world again? It seems to me that giving priority to concept-possession does not get rid of the Veil of ideas; rather, it reinstates the Veil of ideas. Externalism would have seemed to be the cure; not epistemic capacities.
My second point has to do with Frege's Puzzle. Now, the Pragmatist view of concepts gives priority to concept-possession. As I mentioned earlier, making a primitive concept sensitive to the complex contents of beliefs of which it might be a constituent may itself be taken to be inspired by Fregean considerations. It's no surprise, therefore, if the priority of concept-possession (over concept-individuation) to which Jerry strongly objects - courtesy of his Semantic Atomism - offers a natural response to what is called Frege's Puzzle, namely the problem of how a rational thinker S may accept it, or believe, that Fa and not accept it, or believe, that Fb when "a" and "b" are primitive and coreferential, without contradicting himself or herself. Fregean modes of presentation were introduced precisely in order to allow S to rationally believe that Fa (under the "a"-mode) while disbelieving that Fb (under the "b"-mode).
Now, it seems to me, on Jerry's informational line, "Fa" and "Fb" have the same content. Of course, one might try and concoct some notion of narrow content such that for S, "Fa" and "Fb" have the same broad content and different narrow contents. But I take it, Jerry has now given up on this line. So Jerry must - as I have heard him do in lectures a year or so ago (in Paris where he delivered the Jean Nicod lectures) - account for Fregean cases in non-semantic terms, i.e., in purely causal, syntactic or computational terms. On Jerry's view, syntactic structures in the language of thought play precisely the role of Fregean modes of presentation. And I think it's a serious problem to give up on an intentional explanation of what is going on in Fregean cases. The reason it's a serious problem is that giving up on an intentional story seems not compatible with the assumption that S - the thinker - is rational. Not to know that a = b is a case of ignorance, not irrationality. And non-intentional explanations - lapses from intentional explanation - seem to fit irrational, not rational, behavior.
A third (psychological) problem for Jerry's sharp distinction between the semantic and the causal properties of concepts is that it may make it difficult for him to recognize the difference between the contents of genuine propositional attitudes and the contents of merely information-carrying subdoxastic states. I am not questioning the fact that information-carrying subdoxastic states may, as well as beliefs and desires, represent and misrepresent the world. And Jerry certainly has an account of the representational and misrepresentational powers of such information-carrying states, in terms of the Asymmetric Dependency Condition. The (psychological) problem rather has to do with the fact that, I take it, a natural constraint on the conceptual content of beliefs - as opposed to the non-conceptual content of subdoxastic states - is that the causal powers of the former should depend on their contents. In the production of intentional behavior - behavior done for reasons -, the contents of the states having causal efficacy should be doing some job. As many people have pointed out, it is one thing for a state having semantic properties to be causally efficacious in virtue of its non-semantic properties. It's another to be causally efficacious in virtue of having the semantic properties that it has.
Now, Jerry might want to distinguish between subdoxastic information-carrying states and beliefs without adverting to any difference in their semantic properties. He might say: the difference resides entirely in the different causal or functional properties of the vehicles. The difference would be this: in virtue of their causal properties, beliefs belong to a belief-box in which they have the computational potential to interact with any other beliefs (and with other propositional attitudes for that matter). Subdoxastic information-carrying states are modular: in virtue of their computational causal properties, they belong to a vastly smaller box with a vastly more restricted potential for interacting with other information-carrying states. He would, therefore, claim that the difference I am alluding to can be exhaustively accounted for in terms of the causal properties of representations - without appealing to their semantic properties.
But then on the one hand, he has to admit that concepts are partly individuated by their causal properties. Actually, when you think of it (as mentioned above), it's not clear that Jerry is really entitled to his separation between semantics (or metaphysics) and psychology. After all, his own solution to the problem 'Where do mental representations derive their power to represent things from?' appeals to the Asymmetric Dependency Condition, which in turn is no other than a higher order relation between causal dependencies. Now, if mental symbols derive their semantic properties from a higher order relation between causal dependencies, then the divorce between semantics and psychology - or between the semantic and the causal properties of mental symbols - becomes problematic. In general, it seems to me, informational semantic theories must appeal to some causal properties of the symbols whose semantic properties they try to account for (or to some of the properties in virtue of which such symbols enter causal relations).
On the other hand, it seems to me, a conceptual representation of Xs (say, dogs) must differ from a perceptual representation of a particular X (a particular dog), not only in virtue of its causal properties, but also in virtue of its semantic properties, in virtue of its representational qualities. The perceptual representation will be informationally far richer than the conceptual representation. I do not want to get into the issue of innate concepts, which is a question about the origins - the ontogeny - of concepts. I just want to claim that my concept of a dog must be less informationally profuse than my percept of a particular dog. The former, not the latter, must apply to any number of dogs. Being more or less informationally profuse is a semantic property of a symbol.
Fourthly and lastly, I want to go back to the question I raised at the beginning: What is the difference between thinking about Xs and being able to reason about Xs? Consider again Jerry's contrast between giving priority to concept-possession and giving priority to concept-individuation. On the Contemporary view, on which priority is given to concept-possession, "having a concept is having certain capacities. To have the concept of X is to be able to recognize Xs and/or be able to reason about Xs in certain ways". But on the Classical view, on which priority is given to concept-individuation, "having the concept of X is just being able to have thoughts about Xs". As Jerrys says of the latter: "the concept DOG... is the concept of dogs; it's the concept that you use to think about dogs with".
But now, on the one hand, pursuing the previous contrast between percept and concept (between a perceptual and a conceptual representation of a dog), there is a sense in which a concept of a dog - a genuinely cognitive structure - must have to do with recognition or identification of dogs, in a way in which having a perceptual sensory representation of a dog or experiencing a dog does not. After all, some creatures which experience dogs may well lack a dog-concept; and the difference must reside inter alia in what they may or may not do with their dog-representation - in particular, in whether they may or not categorize or sort particular dogs as members of the category of dogs.
On the other hand, the difference between using a concept of X to think about Xs (which, on Jerry's view, is OK) and being able to recognize Xs and reason about Xs (which is not OK) seems to me to be vanishingly small. Any account of using a concept of X to think about Xs will, I surmise, come very close indeed to recognizing Xs and reasoning about them. Again, the gap between semantic and causal properties of concepts seems narrow indeed, as does the contrast between concept-individuation and concept-possession.
 I am grateful to Nenad Miscevic for conversation on the topic of this paper. I am grateful to Dan Sperber and especially Paul Horwich for very helpful criticisms.