More on Modularity

I just read an interesting paper by Musolino and Landau (M&L) on William’s Syndrome (WS) and its implications for the modularity of FL (“Genes, language, and the nature of scientific explanations: The case ofWilliam’s syndrome”). WS is a genetic disorder that has been relatively well localized (“about 25 genes missing from chromosome 7” (124)). As M&L describe it the disorder consists in “a highly unusual cognitive profile, with individuals showing severe impairments in a range of spatial functions but strikingly fluent and well-structured language (124).”  Not surprisingly, WS is “often cited as a compelling demonstration of the modularity of mind (124).” 

For reasons that elude me, the idea that minds have a modular structure seems extremely irritating to some.  What makes this hard to understand is that the standard brain mapping enterprise (the main current activity of neuroscience (see here)) seems to presuppose that the brain is divided into different sectors, with different properties, doing different things. This sure sounds like a commitment to something very much like modularity.  However, given the relation between modularity and domain specificity, the usual suspects have been pushing back against the idea that WS argues for modular minds/brains.  I am no expert in these areas, but I recommend the M&L paper for clarifying the ins and outs of the “debate.” Note the scare quotes. What is most interesting about their reconstruction of the discussion is the disinclination of the non-modularists (aka ‘neuro-constructivists’) to actually engage with linguistic data.  Here’s what I mean.

M&L review work that seems to indicate the WS kids function pretty much like we do in rather complex linguistic tasks.  Not identical but pretty much.  Indeed, there are two basic facts when one compares WS kids to “typicals.”  First, both WS and typicals do well above chance in a series of rather complex scope experiments involving negation and disjunctive ‘or’ (136-7). Second, typicals of the same “mental age” as WS kids do better than the WS kids while younger typicals and WS kids function more or less the same. M&L argue, reasonably it seems to me, that both facts need accounting for.  And they provide an explanation (roughly, that the high success rate of both WS kids and typicals stems from having roughly the same kind of competence, while the differences stem from lesser processing capacities in WS kids).  I found the account very plausible, but that’s not what I want to highlight.

What I want to highlight is the following difference: whereas M&L try to offer an explanation for the two facts noted above, the anti-modular neuro-constructivists never seem to offer an account for the first fact.  Rather than provide an analysis explaining how the kids clearly do what they do, they point to the profile differences and conclude that WS kids must be using different mechanisms. What mechanisms? Well, M&L quote Thomas and Kramiloff-Smith who citing (of course) Christiansen & Chater and Rumelhart & McClelland suggest the following: “presumably …lexical or semantic/pragmatic compensatory mechanisms … that contain some but not all of the grammatical properties outlined in a generative theory, or … computational mechanisms that approximate formal syntactic systems under some processing conditions but not others (139).” Whew! This is to hand waving what a piper cub is to a jumbo jet. Wave any faster and wind power would become our answer to global warming!  In other words, the above sentence is pure, unadulterated poodle poop. M&L are more generous and observe that there is a bit of a gap (“a non-trivial task”) between these insinuations and anything resembling an explanation of the observed phenomena (c.f. 140). M&L are clearly very nice people.

I have no idea whether M&L’s own proposed explanation is correct. Given how interesting the implications of WS are for the modularity issue, it is reasonable to demand high standards of evidence (Why? Because the more interesting a conclusion the more evidence we should demand).  However, at least M&L seem to be playing a game we can all recognize.  The neuro-constructivists seem not be playing at all.

Why don’t the anti-modularists feel compelled to provide a detailed account of the linguistic data?  Here’s my totally uninformed hunches:

First, I suspect they don’t really consider linguistic data to be serious data. It’s not as sexy or impressive sounding as “plasticity, adaptation, interactivity, and lexical or semantic/pragmatic compensatory mechanisms (139).” The fact that we know something about the grammar underlying these phenomena while we know next to nothing about these other factors should not blind us to how pale and wan ‘syntax’ looks next to the muscular sounding four horsemen of neuroscience above noted.

Second, the neuro-constructivists are just sure that modularity must be wrong. After all it implies domain specificity and as any decent empiricist knows (and these people have clear empiricist sympathies) this is impossible. I discussed a version of this position in the last post (here) and here it pops up again.  I find this deep bias against modular-domain specific knowledge quite incomprehensible, especially for anyone who watches Animal Planet or reads the Tuesday Science Times. Hourly we find out that animals have the most incredible domain specific knowledge (the latest one being dung beetles, who apparently navigate using the Milky Way).  And if they do, wouldn’t it be downright weird if humans didn’t?  If anything the burden of proof is heavily on those who resist any modularity in humans for it relies on an odd kind of biological dualism; one that sequesters human cognition from the rest of mammalian psychology. 

M&L have done the rest of us a service. Whether they are right or not about WS and modularity (my money IS on them), arguing against it requires explaining all the data, the linguistic data included. Those that don’t do so are not even wrong. They are irrelevant.