Notes
- A somewhat negative review of "Karmiloff-Smith (Annette) - Beyond Modularity".
- I was interested in Modularity of Mind, Innateness and Connectionism in my undergraduate Philosophy of Psychology course at Birkbeck back in 2002 or thereabouts. I've not really touched on it since, I don't think. I should review – and convert – my undergraduate Essay on Modularity of Mind1
- Anyway, I've now read this Paper, but it'll make more sense once I've read the book that it reviews. Indeed, if I ever get the time, it’ll be good to iterate re-reads of both.
- Basically, Fodor is unimpressed by Karmiloff-Smith’s ‘RR Thesis’ (see below). He thinks that any ‘theories’ older children (and adults) concoct in domains covered by Modules are made on the basis of output from these modules rather than changes to the modules themselves.
Introduction
- These days, hordes of people are interested in the idea that aspects of cognitive architecture may be modular. I know at least two or three (people, not hordes or aspects), and there may be others. But "modularity" means different things on different tongues. In this chapter, I want briefly to distinguish between some versions of modularity theory that are currently in play. Then, I will discuss one of them in critical detail.
- There are four essential properties connected with the notion of a module: Unless you believe that at least some mental entities instantiate at least two of them, you are not in the modularity camp according to my way of choosing sides.
- Encapsulation: Information flow between modules — and between modules and whatever unmodularized systems the mind may contain — is constrained by mental architecture. "Constrained by mental architecture" means "not cognitively penetrable" (Pylyshyn, 19842): You can't change such an arrangement (just) by fooling around with someone’s beliefs and desires. In particular, architectural arrangements are (relatively) insensitive to instructional variables in experimental tasks. The persistence of illusions is the classical instance. Convincing the subject that the Muller-Lyre effect is illusory doesn't make the apparent difference between the length of the lines go away.
- Inaccessibility In effect, the inverse of encapsulation. Just as information about beliefs and desires can't get into a module, so the information that is available to its computations is supposed to be proprietary and unable to get out. In particular, it is supposed not to be available for the subject's voluntary report.
- Domain specificity The information and operations by which a module is constituted apply only in the module’s proprietary domain. Concepts and processes may thus be available for language learning, or face recognition, or space perception which are not likewise available for balancing one's checkbook or deciding which omnibus to take to Clapham.
- Innateness The information and operations proprietary to a module are more or less exhaustively “genetically preprogrammed" (whatever, exactly, that means).
- Table 11.1
| | NC | AKS | JAF |
| Encapsulated | Don't care * | Yes and No # | Yes |
| Inaccessible | Yes | Yes and No % | Yes |
| Domain-specific | Yes | Yes | Yes |
| Innate | Yes | Yes and No @ | Yes |
- Note: NC is Chomsky, AKS is Karmiloff-Smith, and JAF is me.
- People who agree that some mental processes are modular may, nonetheless, differ appreciably in their views about the encapsulation, accessibility, domain specificity, and innateness of even their favorite candidates.
- Table 11.1 shows a rough sketch of the way three currently active theorists distribute, all of whom think of themselves as pro-module in some sense or other. Comments:
- *: Chomsky, in some of his moods, dislikes the whole information- processing view of mental operations. If the mind isn't an information processor at all, then the question of whether it's an encapsulated information processor doesn't arise.
- #: A proposed principle of the ontogeny of cognition: Mental processes become encapsulated in the course of cognitive development (perhaps through overlearning). So they are encapsulated synchronically but not diachronically. I'll refer to this putative process as modularization (my term, not Karmiloff-Smith's).
- %: Another proposed principle of cognitive development: Modularized information becomes increasingly accessible over time as a result of an "epigenetic" process of representational redescription (the "RR" theory3). I'll refer to this as a process of demodularization (again, my term).
- @: What's innate: Some domain-specific information and "attentional biases"; and, presumably, the psychological mechanisms that underlie the putative epigenetic processes. But neither encapsulation nor accessibility are themselves genetically preprogrammed.
- An aside about attention: It's a recurrent theme in Beyond Modularity (BM) and also in Elman et al. (see chapter 12) that "There must be some innate component to the acquisition of [e.g.] language [but] ... this does not mean that there has to be a ready-made module. Attention biases and some innate predispositions could lead the child to focus on linguistically relevant input and. with time, to build up linguistic representations that are domain specific" (36). This emphasis on innate attentional biases is not widely shared by modularity theorists. It strikes me as unpromising, and I won't discuss it in what follows. In neither Karmiloff-Smith's book nor Elman's is it explained how one could have a disposition (innate or otherwise) to concentrate on Xs unless one already has the concept of an X. ("Pray, attend to the passing flubjumbs." "Can't.” "Why not?” "Don’t know what a flubjumb is.") Postulating innate attentional biases doesn't dispense with the postulation of innate conceptual content; it just presupposes it.
- It may be that, in passages like the one just quoted, Karmiloff-Smith is only suggesting that it would be a help to the child to be (differentially) interested in speechlike sounds. That's quite plausible, in fact; but it doesn't even begin to explain how someone who is so biased manages "with time, to build up linguistic representations that are domain specific." As far as anybody knows, you need innate conceptual content to do that; indeed, as far as anybody knows, you need great gobs of it. (I am disposed to attend to the speech sounds that German speakers make; but I find learning German very hard for all that.)
- So much for some current kinds of modularity theories. Perhaps I should say at the outset that I think you'd have to be crazy to bet much on which, if any, of them is true. The study of mental architecture is in its infancy, and it looks to be developing very slowly. My modest ambition in what follows is just to indicate some doubts about Karmiloff-Smith's view. And, even here, I'm not going to argue for anything so positive or decisive as that she is plain wrong about modularity. I will, however, try to show that the ways she sets out her view, and the ways she undertakes to defend it, are insensitive to certain distinctions that a cognitive architect really ought to observe. And that, when this is all cleared up, what’s left may after all be true — who knows? — but, as things stand, neither the arguments for modularization (the thesis that cognitive architecture becomes increasingly modular with development) nor the arguments for demodularization (the thesis that information in modules becomes increasingly accessible with development), are persuasive.
Conclusion
- Where does all this leave us? There are two issues that everybody is trying to get clear on: What is the cognitive architecture of the adult mind? and, whatever it is, how did it get to be that way? I think the empirical results over the last couple of decades make a not implausible case that modular architecture is an appropriate idealization for some aspects of the organization of mature cognition. About the second question, however, nothing is really known; we're all just playing our hunches. Perhaps the deepest issue that divides people in the theory of cognitive development is whether there are ontogenetic processes that affect cognitive architecture. Karmiloff-Smith and Piaget are betting that there probably are; Chomsky and I are betting that there probably aren't. All I can tell you for sure is this: There may in fact be architectural changes in the course of the ontogeny of cognition, but nobody has found a clear case of one so far — and since the developmental plasticity of the mind has been a main theme of Anglo-American psychological speculation for a couple of centuries, I do think that fact is striking.
Comment:
In-Page Footnotes
Footnote 2: Footnote 3:
- I’m not as sure as I'd like to be about what an epigenetic process is. Karmiloff-Smith provides the following gloss on Piaget's usage: 'For Piaget both gene expression and cognitive development are emergent products of a self-organizing system that is directly affected by its interaction with the environment" (9). But a self-organizing system that is, nevertheless, "directly affected by its interaction with the environment" is, to my taste, rather like a circle that nevertheless has comers. Probably "epigenetic" just means "not primarily input driven." I shall assume that is what it means.
Text Colour Conventions (see disclaimer)
- Blue: Text by me; © Theo Todman, 2025
- Mauve: Text by correspondent(s) or other author(s); © the author(s)