Splitting Concepts

A widespread assumption in the literature on concepts is that concepts form a unified, or singular, natural kind, as opposed to a set of different natural kinds. That is, advocates of different accounts disagree on which kind of thing constitutes concepts, but they agree that there is only one such kind of thing (with different instances for each category).

In recent years, two young philosophers, Dan Weiskopf and Edouard Machery, have completed Ph.D. dissertations questioning this assumption, independently of each other. Dan graduated from Washington University in St. Louis roughly at the time I went there for my PNP postdoc (2003). I met him and learned about his work.

In the fall 2004, I co-taught a course on concepts with Sam Scott, who at the time was the other Wash U PNP postdoc. Sam did some ground-breaking empirical work on non-referring concepts, so he knows a lot more than I did about the psychological literature on concepts.

I wanted to learn more about Dan’s work and discuss it in class, so I asked Dan for a copy of his paper in progress, which is now available on his website. Unfortunately, the paper may not have been ready at the time. At any rate, he didn’t send it to me. But during the course, Sam and I discovered Edouard’s paper, Concepts Are Not a Natural Kind, which at the time was forthcoming in Philosophy of Science (it is now published).

Since we didn’t have Dan’s paper, we decided to read and discuss Edouard’s paper instead. We were not entirely satisfied with Edouard’s argument, but we liked the conclusion—concepts split into different natural kinds—and we thought we could find better arguments. So we wrote a paper of our own.

Our paper, entitled “Splitting Concepts”, is now officially forthcoming in Philosophy of Science.

In our paper, the argument I most care about is what we call the “argument from language”. Its conclusion is that there are at least two different kinds of concept, which may be called linguistic concepts (those that explain linguistic abilities) and non-linguistic concepts (those that explain cognitive abilities we have in common with non-linguistic animals and babies). I think this conclusion may have some appealing philosophical payoff:

Many philosophers like to point out that there are dramatic cognitive differences between linguistic and non-linguistic creatures. Language seems to carry with it a lot of cognitive power. If you combine this observation with the two common assumptions that linguistic abilities are explained by concepts and that concepts are a singular natural kind, you face a dilemma. Either non-linguistic animals and babies have concepts or they don’t. If they do, you should explain why they are not as smart as we are even though they have concepts. If they don’t, you should explain why they are as smart as they are even though they have no concepts. Either way, you need a different theory of cognition for non-linguistic animals and babies as you do for linguistically competent human beings. Needless to say, philosophers have explored both options, but I’ve never been satisfied with either of them. Alternatively, you could try to downplay the cognitive differences between linguistic and non-linguistic creatures, but that doesn’t sound very appealing either. Now there is a better way out: reject the assumption that concepts are a singular natural kind. Animals and babies have one kind of conceptual mental representations. Then, babies develop a new kind of conceptual representation—linguistic representations. Once linguistic representations are in place, humans are well on their way to surpassing the intelligence and inferential power of other animal species.

Of course, this is little more than a slogan at this point. Perhaps some day it will turn into a theory.


  1. Anibal

    I´m totally agree with yours (at least at some point) and Machery and Weiskopf line of thought about concepts and why they have to accomadate the vast and distinct kind of information (visual,propioceptive, acustic, abstract..) organized in the brain in terms of variety of representational formats and not a single one.
    I think that cogntive activity is not exhausted by linguistic actvity ( classically concieve “conceptual organization”), although language boost cognitive functions to diferentiated levels.

  2. Edouard Machery

    A short reply.

    First on anibal’s comment. It is important to realize that the notion of concepts does not pick out *any* knowledge stored in long-term memory or used in our cognitive processes. Rather, the notion of concept picks out a specific subset of the bodies of knowledge stored in LTM or used in our cognitive processes. Thus, from the fact that we have plausibly many different kinds of information in LTM, it does not follow that the class of concepts is not a natural kind.

    Second, on Gualtiero. It won’t come as a surprise that I disagree with Gualtiero and Scott’s evaluation of my argument. In the draft I read, the main objection was that if I were correct, I would not have shown that concepts are not a natural kind. Rather, I would have provided an argument for some kind of hybrid theory of concepts. Hybrid theories of concepts, for memory, contend that concepts divide into parts that are involved with different cognitive competences (for instance, categorization vs. concept composition or fast categorization vs. slow, reflexive categorization) and encode different types of information.

    As I will explain at length in my reply to their papers, which should come out in Philosophy of Science with Gualtiero’s paper or in another issue, as well in my forthcoming book Doing without Concepts (Oxford UP), this objection is unconvincing.

    A few preliminary thoughts:

    First, if I were right, it would not the case that the different kinds of concepts are involved with different cognitive competences. Rather, I contend, and evidence suggests, that these kinds of concepts are all involved in categorization, in concept composition, and so on. Importantly, they are not involved in the same cognitive processes. For instance, we have different cognitive processes for categorizing, which use different types of concepts. These cognitive processes sometimes yield the same output (i.e., the same judgment of categorization), but sometimes do not. A large body of evidence supports this claim (reviewed in my dissertation and in my forthcoming book). My views have thus little to do with the standard hybrid theories of concepts.

    Second, Gualtiero and Sam Scott do not discuss the large body of evidence that suggests that we do have some coreferential prototypes, exemplars and theories. This is puzzling. The empirical research supports this claim, even though psychologists have for a long time failed to realize that this was the best reading of the evidence (this is were Dan and I agree). A large part of my project consists in drawing the conclusion that this fact suggests that the notion of concept has little value for a scientific psychology. Instead of discussing the evidence for coreferential concepts that belong to very different kinds, Gualtiero and Sam scott propose an entirely speculative argument based on unverified empirical premises. This is a puzzling strategy.

    Anyway, more to come in my reply for PoS.

  3. gualtiero piccinini

    I look forward to reading the details of Edouard’s reply to our paper (and his book). In our paper, we argue that the evidence offered by Edouard in his paper doesn’t go far enough, and that to make a convincing case, he needs precisely the kind of evidence he mentions in the above post. Glad to see he is going to fulfill our request.

    His argument is not imcompatible with ours. Concepts may split along several fault lines. Edouard and Dan have been exploring one. In our paper, we (briefly) discuss two other possible ones, and we do indicate what evidence we think there is for them.

  4. Daniel Weiskopf

    Also a short reply:

    One difference worth noting between myself and Edouard is that I do think that concepts have a robust role to play in psychological theorizing. I don’t tend to put the issue in terms of “natural kinds”, but I prefer to talk (I hope a little more neutrally) just in terms of kinds, understood in as metaphysically harmless a way as possible. So I think that concepts are a superordinate cognitive kind that happens to subsume a variety of distinct representational subkinds. These subkinds get unified by the fact that there are cognitive processes and mechanisms that operate over them in a way that is indifferent to the particular representational subkind that they belong to. Finally, I think that psychologists have traditionally assumed that concepts will be constitued by a single representational kind; here Edouard (and Gualtiero & Sam) and I agree that the evidence suggests otherwise.

    I also don’t hold out much hope for hybrids, at least as they are standardly understood. There is actually some interesting to work to do in distinguishing hybrid models from the sort of pluralist model that I favor. Hybrids tend to face two major sorts of problem. First is the problem of bloat: as we discover new sorts of information and new information-carrying structures that are tapped in new tasks, must we include them in our hybrid concepts? If so, then they will rapidly become too unwieldy to serve as plausible vehicles of thought (on one understanding of what is required to be a genuine vehicle of thought).

    A more serious concern is that at a sufficiently coarse grain of analysis it isn’t the case that “tasks” line up perfectly with representation types. So I think that the evidence cuts against the idea that there are tasks that invariably use a particular type of conceptual representation. You can always fuss around with the boundaries of the “tasks”, of course, but this seems ad hoc.

    On Gualtiero & Sam’s idea that nonlinguistic creatures have one sort of concept and linguistic creatures have another, I’m also skeptical that the linguistic/nonlinguistic barrier is what divides concepts into separate kinds. For one thing, the existence of many kinds of concepts has been demonstrated using both linguistic and nonlinguistic materials, and with prelinguistic infants (and, more controversially, animals) as well as with language-using subjects. Even infants display the ability to use prototypes, exemplars, and rudimentary causal models, for instance. So the evidence that this is the difference that language makes seems thin to me.

    There’s lots more to say about the interface between the conceptual and linguistic systems, and also about “why we are so smart” (to borrow Gentner and Spelke’s expression) but it’s best saved for later, I think.

  5. Edouard Machery

    Dan writes: “These subkinds get unified by the fact that there are cognitive processes and mechanisms that operate over them in a way that is indifferent to the particular representational subkind that they belong to.”

    This is indeed what distinguishes Dan’s views from mine. I doubt the empirical evidence supports this claim. I am curious to know what Gualtiero thinks of this issue.

    I agree with Dan’s point that much of the empirical evidence for prototypes, for theories and for exemplars has been found with both linguistic and non-linguistic stimuli. I take this to be a problem for Gualtiero and Sam’s views.

    Finally, it is worth noting a strange fact. We all agree that there is something like a consensus in psychology about the homogeneity of the class of concepts. But, as Dan told me, at least several psychologists deny this. Self-denial?

  6. Dan Ryder

    Hi Gualtiero,

    Glad to see that my SPP comments proved useful for the new version of the paper – I’m particularly glad to see that you are allowing for the possibility of an overarching kind here (as Dan W. thinks, and I agree, more or less). I don’t think you shored up your defense of the linguistic/non-linguistic fault-line well enough though (and I see Edouard and Dan W. are putting pressure on that too). My original worry was that your arguments to the effect that linguistic competence requires a special kind of representation applied only to syntactic representations. In response, you seem to be saying that the traditional category of “concept” was meant to include syntactic representations all along, and that it is a mere “terminological accident” that psychologists and linguists don’t call syntactic representations “concepts”. A short quote from the new version of your paper:

    “There is little doubt that [desiderata] (3) to (6) require linguistic competence, that linguistic competence requires syntactic competence, and that syntactic competence requires special kinds of representations, which are likely to differ in kind from those of other cognitive processes. We are merely pointing out that these statements, which are hardly controversial, have consequences for the theory of concepts—namely, the theory of the representations that explain phenomena (1) to (6). The fact that psychologists and linguists do not call syntactic representations conceptual may help explain why they haven’t noticed these consequences, but it does nothing to diminish them.”

    First: surely the psychologists’ and linguists’ use of a different term here indicates that they think syntactic processing involves a different kind of representation, i.e. some non-conceptual kind of representation – and isn’t the paper supposed to be about the scientists’ notion of concepts? And second, the theory of concepts is not “the theory of the representations that explain phenomena (1) to (6).” There’s no such thing as *the* representations that explain (1) to (6). (1) to (6) will involve all sorts of perceptual representations, for instance, and most scientists doubt those belong to the same kind of representation that concepts do. (Note that many think that syntactic representations are more like perceptual representations than conceptual ones.)

Comments are closed.

Back to Top