Domain Specificity Revisited: Response to Keil
by Fiona Cowie
California Institute of Technology
When I wrote What’s Within (Cowie 1999), part of the aim was to provide a characterization of nativism and empiricism that explained why nativists and empiricists took (and still take) themselves to be offering radically opposing answers to the question whether any ideas are innate. The problem, I felt, with earlier understandings of the nativism issue was that they didn’t make sense of its being a debate. The idea, for example, that nativists were asserting and empiricists denying that our ideas are innate in the sense that they come ‘from the genes’ rather than ‘from experience’ doesn’t make sense of the nativism controversy. For while asserting that our ideas come from the genes alone is surely a good way of exciting empiricist ire, it’s not the kind of thing that any nativist in her right mind would or did say. And while asserting that ideas come from the genes more than from experience is certainly a thing that sensible people might think, it’s hard to see why empiricists would get all worked up about it, no-one, after all, having the remotest clue how to quantify the respective contributions of the genetic and the experiential to our mental lives. Standard interpretations make nonsense of the nativism-empiricism controversy, I argued, either by making the parties to it come out saying things that are just plain stupid, or by making them say things that their opponents don’t contest.
On the basis of an examination of the two different kinds of argument that nativists and empiricists used to support their respective positions – “impossibility arguments” (IAs) and “poverty of the stimulus arguments” (POSAs) -- I maintained that the nativism-empiricism controversy was actually two controversies run together, and that it concerned the following two questions:
I maintained that this interpretation did not suffer from the defect just adumbrated. For in each of these cases, I argued, something important and substantive is at issue and it’s thus clear why nativists and empiricists should have taken themselves to have been in disagreement with one another. After all, the question whether the processes by which minds are furnished are susceptible of natural scientific investigation is one that reasonable people are still debating today. And the question whether the psychological mechanisms responsible for the acquisition of a certain competence or body of knowledge are task specific or domain neutral is likewise still of prime importance in psychology and cognitive science.
In his thoughtful and thought-provoking comments, Keil argues that on the second of these points, at least, I am wrong. Viewing nativists and empiricists as clashing over the domain specificity or neutrality of our psychological mechanisms gives us no neat taxonomy of positions or views, frequently leads us to mis-classify nativists as empiricists (and vice versa), and generally fails to make sense of their controversy. Far from being a clear dispute over a well-defined question in theoretical psychology, the nativism-empiricism debate instead inherits the murkiness and relativity inherent in the notions of domain specificity and domain neutrality. Absent some clearer specification of what domain specificity (and neutrality) are supposed to be, my suggestion that we view nativists and empiricists as disagreeing on this score provides scant illumination.
1. Keil’s Complaint
Keil’s strategy in ‘Nurturing Nativism’ is the mirror of my own in Chapter One of What’s Within, where I criticize other interpretations of the nativism debate. In essence, what he does is to offer (on my behalf) a number of precisifications of the notion of domain specificity, arguing that since none of them is the right notion to capture what’s at stake between nativists and empiricists, my interpretation “isn’t enough to distinguish nativism from empiricism.” (Keil) His points are as follows:
2. Domain Specific Processes and Innateness
Keil, I think, is both right and wrong here. He’s right that in What’s Within, I did assume without argument or even serious thought that the domain specific/domain neutral distinction is unproblematic. He’s right too that this assumption is unjustified. Not only did I fail in the book to provide any explication of what the distinction consists in, I now have my doubts (and for much the reasons Keil elaborates) that such an explication is in general possible.
However, the general unclarity of the notion of domain-specific processing need not undermine my interpretation of the nativism controversy, so long as it is possible to distinguish domain-specific from domain-general learning. For what I claimed in What’s Within is not that nativism commits you to domain-specific psychological processing tout court, but rather that it commits you to domain-specific learning processes. Nativism, after all, is a broadly developmental view about learning (or more broadly the acquisition of knowledge and skills). It’s not a global view about the nature of psychological processing (i.e., whether it’s domain specific, or ‘modular,’ or not) in general. If they’re committed to a view about domain specificity, then, it stands to reason that nativists will be committed to a view about the domain specificity of the processes of learning (/acquisition) – not (or not in the first instance) to a view about the specificity of the many other non-learning processes which Keil discusses – object recognition, vision, face recognition, theory of mind, etc. So as long as it’s possible to say what a domain specific learning mechanism might be, and as long as it’s possible to do so in a way that (a) gets nativists and empiricists categorized correctly and (b) gives them something substantive to argue about, my interpretation of the nativism-empiricism controversy is not undermined by Keil’s arguments – which I am pretty sure are correct – that it’s not in general useful to think about psychological processing in those terms.
3. Domain Specific Learning Mechanisms
All that’s required to make sense of the nativism-empiricism controversy is an account of what makes a learning mechanism more or less domain specific. We do not, contra Keil, need an account of domain specificity (or “modularity”) that successfully categorizes all sorts of psychological processes as being general or specific. I used to think – and my failure to address this issue in What’s Within is a reflection of this optimism – that it would be a fairly trivial matter to give an account of domain-specific learning. Now, however, having given the matter more thought, I’m not so sure that the domain specific/domain neutral distinction is of much use, even as applied in this limited arena.
In order to develop and account of a domain-specific learning mechanism, we need to do two things. First, we need to come up with a principled way to individuate domains. Second, we need to say what makes a learning mechanism specific to a domain (or not).
Neither of these tasks is particularly easy. On the question of what constitutes a domain, Keil argues that a domain must be “more than any pattern of information in the world for which the body has a specialization.” For, as he points out, it doesn’t follow that just because we are innately allergic to poison ivy, we have a domain-specific mechanism for learning about poison ivy, “even if those allergies result in all of us learning about poison ivy differently from daisies.” (Keil) In addition, it’s clear that we cannot rely on our intuitions as to what constitutes the domain of some learning process: our pretheoretic ideas in this case (as in others) might be misleading. We might think, for example, that “language” is obviously a natural domain, thus expecting a single acquisition mechanism to account for the acquisition of linguistic competence. However, we must acknowledge the possibility that a number of different mechanisms with different domains might collaborate to produce linguistic competence in us. Our views about what the domain of a learning process is must be revisable in the light of what is discovered about how that process actually works and interacts with other processes.
Even taking these admonitions on board, however, delineating a learning mechanism’s domain remains problematic. Suppose, for instance, that the domain of a learning mechanism is taken to be whatever range of information the mechanism is able to assimilate. This conception is clear enough, I suppose, but it will not support our drawing a further distinction between domain-neutral and domain-general learning processes. That is, if the domain of the learning process is simply whatever range of facts or situations that process allows you to learn about, then not only does every process have a domain, but every process is “domain specific”! If operant conditioning, for example, underpinned our learning how to sex chickens and our learning how to drive a stick shift, then the “domain” of that process would be chicken-sexing-and-shifting-gears and operant conditioning, having its own proprietary domain, would apparently have to be classed as domain specific. (Of course, this seems like an odd sort of domain for a learning process to have – but recall that we are putting aside our intuitions about what ‘natural domains’ might look like.)
What this shows is that in order to have any hope of distinguishing between domain general and domain neutral processes, we need a way to distinguish the range of facts or situations that a learning mechanism de facto is able to process from those that constitute its ‘natural’ or ‘proprietary’ domain. Then, a mechanism would be domain specific if it only operated in its proprietary domain, and domain general if either it had no proprietary domain or it operated outside that domain.
One obvious way to draw a line between what a mechanism happens to do and what it ‘should’ do or what it is ‘natural’ for it to do, is by appeal to its biological function. That is, we could say that a learning mechanism’s natural or proprietary domain is whatever problem(s) that mechanism has the biological function of solving – it’s whatever it evolved to enable us to learn. Thus, if a given learning mechanism, L, was selected for because it enabled individuals to increase their fitness by learning how to speak and understand a language, then “language” is the domain of the mechanism L. If, however, L was selected for some other purpose – say, learning how to detect regularities in the non-verbal environment – and only subsequently was co-opted for use in learning language, then the domain of L is not “language” but “statistical regularities.”
On the present suggestion, a nativist about (say) language is committed to the following two ideas:
(i) There is a learning mechanism which evolved because it enabled us to learn language; and
(ii) This learning mechanism is structured such that it cannot be (or maybe just is not) used for any other learning task.
Empiricists about language learning, on the other hand, would deny (i) or (ii) or both. They might deny (i) and claim that there is nothing which evolved because it enabled us to learn a language. (This, of course, is rather implausible, given the manifest advantages linguistic competence must have bestowed on our ancestors.) Or they might deny (ii) and argue that although there is a mechanism which was selected for because it enabled us to learn language, that mechanism is not specific to that domain: it can be (/is) used for other purposes as well. Or they might deny (i) and (ii) on the grounds that the mechanism which is now used to learn language originally evolved for other purposes and thus has a domain that is not “language”, even though, being domain-general, it is now used for purposes of learning language.
This account has some desirable features. For instance, as desired by Keil and on the plausible assumption that there is no learning mechanism that was selected for because it enabled us to learn about poison ivy, it turns out that there is no learning mechanism whose domain is: poison ivy. Also, on this view, it’s not our pretheoretic ideas about what is and isn’t a “natural” domain that count. Instead, the domain of a mechanism is that class of problems the ability to solve which as a matter of fact made a difference to our ancestors’ survival and reproduction. Finally, this account of domains – unlike the account discussed earlier – does allow us to draw a distinction between mechanisms that are domain general and domain neutral. A mechanism is domain specific if it is used only to learn about the things within its domain. (Thus, if L were still used just for learning about language, it would be a domain-specific language learning mechanism.) And a learning mechanism is more domain neutral, the more things outside its domain it is used to learn about. (Thus, if L were used both for learning about statistical regularities and language, when its domain was just one of these, it would to that extent be domain general.
The present account also gets at least some nativists and empiricists correctly categorized. Someone like Pinker, for instance, comes out a nativist. On his view, the LAD was selected because it enabled us to learn language, and that enhanced our fitness (see Pinker and Bloom 1990, Pinker 1994). Thus, its domain is language. Further, the LAD is a domain-specific learning mechanism because given how it works (i.e., given that it operates via the triggering of the parameterized principles of UG), it is not the kind of learning mechanism that can be used in other domains. Similarly, someone like Skinner comes out an empiricist, although this could be for either of two reasons, depending on how he told the evolutionary story. On the one hand, he might deny both (i) and (ii), arguing that the mechanism of operant conditioning evolved in the first instance to enable us or our ancestors to (say) learn how to act so as to get what we want or need (so that “planning” or “practical reasoning” – not “language” – would be the domain of the learning mechanism) and that that mechanism is also domain-general (since it’s used both in its domain and for language learning as well). Alternatively, however, Skinner might accept (i) and deny only (ii), arguing that the mechanism of operant conditioning was the one which happened to get recruited for the purposes of learning language – its domain, then, would be “language” – only its operations are not specific to that domain, because it so happens that that mechanism can be used for other learning tasks as well. (The second of these strategies would, of course, be implausible, given that the mechanism of operant conditioning is apparently very old, appearing in species which branched off from our family tree well before the appearance of language. But this doesn’t alter the point that in other cases, it might be plausible to think that a mechanism which originally evolved because it could do task A subsequently got used to do other tasks as well or instead.)
4. Residual Problems
I used to think that something more or less like this would do to fill out the notion of a domain specific learning device, and thus to plug What’s Within’s lacunae on this score. Now, however, I am not so sure. One problem with this account is that while domain specificity is well defined, domain generality comes in degrees. That is, we can say that a mechanism is domain specific, period, so long as it is only used for learning within its domain, that is, so long as it only does what it evolved to do. But we can only say that it is more or less domain general: a mechanism is the more domain general the more its operations generalize to other learning tasks. Now, this might not be a problem in principle – lots of perfectly good distinctions, after all, involve matters of degree. But I think it is a problem for me, given the use I want to make of the domain-specific/domain neutral distinction. As stressed in §1 above, I want that distinction to afford a clear demarcation between nativists and empiricists. But while the distinction as currently conceived makes it clear what it is to be a nativist, it leaves the empiricist’s commitments rather vague. Empiricism looks like nothing more than the denial of nativism (”such and such learning mechanism is not domain specific”) and while nativists may be happy with viewing empiricism as nothing more than a denial of their position, empiricists would presumably reject such a characterization, and rightly so.
Another problem with this approach to domain specificity and neutrality is that it glosses over a whole lot of issues that have arisen in discussions of the notion of function in biology. Or, to put it another way, the present conception of the domain of a learning mechanism is indeterminate in whatever ways the notion of a biological function in general is indeterminate. For example, if it’s in principle unclear (as, e.g., Fodor 1990 has argued) whether the biological function of a trait is F as opposed to F-or-G, it will be unclear what the domain of learning mechanisms are, because domains, on this account, are derived from functions. Similarly, if it’s unclear (as a number of philosophers of biology have argued – see, e.g., the papers by Bigelow and Pargetter, Neander, Godfrey-Smith, Walsh and Ariew, and Buller in Buller 1999) what period of our evolutionary past to look to in nailing down the function of a trait, it will be unclear where to look in delineating a mechanism’s domain. Is the function of feathers heat regulation (given that feathers apparently were first selected for because they kept their wearers warm) or flight (given that they were subsequently selected for this purpose)? Similarly, is the domain of Pinker’s LAD really language? Or is it, say, aimed throwing or tool-building, given that the neural structures recruited for language learning have been argued (by Wilkins and Wakefield 1995 and Greenfield 1991 respectively) to have evolved originally for these purposes? The fact that the history of the human animal appears to be a story of massive and continual exaptation (Allman 1999) – that is, the fact that many traits, including learning mechanisms, can contribute to an organism’s fitness in many different ways over time – means that there is a considerable amount of slippage in the question of what their function (hence their domain) might be. Again, this is potentially damaging to my account.
Finally, and this is for me the most problematic (though also perhaps the least interesting) feature of the present picture of domain specificity and domain neutrality, it is surely implausible to attribute views about the evolution of learning mechanism to historical nativists such as Leibniz and Descartes. Even Chomsky would resist being attributed a view about how the learning mechanism evolved. In What’s Within I argued that it’s not too much of a stretch to attribute views about special-purpose learning mechanisms to the Dead Guys. But if that is tantamount to attributing to them views about the evolution of those mechanisms, then that is clearly a problem. It’s unclear to me how much conceptual revision it is possible to do before one is no longer interpreting past thinkers, but rather rewriting them. Still, I suspect that if this is how the notion of domain specificity (and neutrality) are to be cashed out – and I confess I can’t see any other plausible approach – then I may have gone too far in What’s Within.
In pressing me on the issue of domain specificity, Keil has accurately put his finger on what I see as the major flaw in What’s Within. As indicated here, I am no longer sure whether the notion of a domain-specific learning device (let alone the notion of domain-specific psychological processing in general!) is salvageable. What if it’s not? Where does that leave nativists (and What’s Within)? In conclusion, I would like to make two points.
First, if the notion of a domain-specific learning device really is hopelessly confused, then there are two possible conclusions we might draw. On the one hand, we could conclude that I got nativists wrong when I thought that they were concerned with special-purpose learning devices. On this view, the interpretation must be wrong because people like Descartes, Leibniz and Chomsky are way too smart to be talking about something so obscure. On the other hand, though, we might still be convinced on the basis of the exegetical and dialectical considerations adduced in What’s Within that nativists were in fact concerned with domain specificity, and conclude, perhaps that because that notion is so obscure, they shouldn’t have been. I know which way I would go on this one, of course, but no doubt others have their own ideas.
Secondly, I would point out that none of this undermines my identification of a second strand in nativist theorizing, the ‘mystery mongering’ or non-naturalism engendered by Impossibility Arguments. It seems to me that a kind of explanatory pessimism is an element in the thinking of many nativists, and that this pessimism is in significant part what empiricists object to about nativists’ accounts of acquisition. So even if I’m wrong about domain specificity’s being an important nativist commitment, I hope that What’s Within’s identification of an anti-naturalist tendency in some nativist thinking still makes a small contribution to our understanding of these deep, dark and difficult matters.
Allman, J.M. 1999: Evolving Brains, New York : Scientific American Library.
Cowie, F. 1999: What’s Within: Nativism Reconsidered, Oxford University Press, New York.
Buller, D.J. (ed) 1999: Function, Selection and Design, SUNY Press, Albany.
Fodor, J.A. 1990: A Theory of Content and Other Essays, MIT/Bradford, Cambridge, MA.
Greenfield, P.M. 1991: ‘Language, tools and brain: The ontogeny and phylogeny of hierarchically organized sequential behavior,’ BBS 14:531-95.
_____ and Bloom, P. 1990: ‘Natural Language and Natural Selection,’ BBS, 13: 707-84.
Wilkins W.K. and Wakefield J. 1995: ‘Brain evolution and neurolinguistic preconditions,’ BBS 18:161-226.