The Adaptable Mind: What Neuroplasticity and Neural Reuse Tell Us about Language and Cognition - John Zerilli 2021

Preliminary Remarks
The Language Module Reconsidered

The contention that language is domain-dedicated and innate is as much a claim about cerebral organization as it is about function. My aim in this chapter is to extend the framework provided so far by offering an account of how language could be implemented in brains in a way that honors its autonomy, developmental robustness, and connection to other domains. Any examination of the relevant literature will quickly dispel the illusion that there can be certainty in a field like this, at least for the present. But there is more than enough evidence, I think, to make the prospects of some proposals doubtful enough to warrant serious skepticism—in particular, the claim that language is subserved by hardwired and dedicated neural circuitry—and enough evidence, too, to provide the basis for a sensible if only tentative conception of neurolinguistic organization. Because all such proposals to date (no matter how vigorously and at times dogmatically defended) have been advanced in a spirit of scientific speculation, my own, of course, will be no different. I intend my thoughts on the subject to count as one further effort in the ongoing attempt to render plausible how something with the particular characteristics of language could be implemented in a domain-general architecture. The need for such a project to succeed has become urgent, in my view, precisely because the alternative is too much at odds with what we do know about the brain. Short of compelling reasons to the contrary, a theory of cognitive architecture should strive to be consistent with as much of the hard evidence that we have at our disposal, be it neural, psycholinguistic, developmental, evolutionary, or computational. But domain-specific accounts of the functional architecture of language can no longer assert that they meet this desideratum.

I should note that, while the framework of reuse I have adopted in the book so far will continue to do work for me in the present context, I shall at this juncture have to part company with Anderson and other proponents of reuse. After much reflection, I have come around to the view that neural redundancy should be assigned a much more prominent role in theories of cognitive architecture than many proponents of neural reuse seem willing to contemplate. It strikes me that, in view of how simple and powerful the principle is, it is a wonder that more has not been made of it in debates concerning the modularity of mind. In my view, this is a significant omission, although happily one that, if remedied, can go a long way towards reconciling the evidence of linguistic modularization and neural reuse. I introduce what I call “the Redundancy Model” in § 7.5.

The chapter proceeds as follows. First up, we need to get a little more clarity on the very idea of a “language module.” What are we looking for? What does it mean to say that language is modular, or represents a cognitive specialization? Any answer presupposes some conception of the language domain as a psychological phenomenon, as well as some conception of specialization at the level of implementation. Regarding the second issue, I should think I have already said enough, so even though I rehearse a couple of competing conceptions later in this chapter, I do not intend to modify the position that has sustained the investigation thus far. Regarding the first issue, however, I have so far said very little. The two most influential conceptions of the language domain are those associated with the linguist Noam Chomsky and the philosopher Jerry Fodor. While it would not be wrong to see these two thinkers as belonging to the same broad school of thought, their conceptions of language—of what it is we should be looking for within a language module—are very different. The evidence I adduce raises problems for any defender of linguistic modularity, no matter where they fall on the Chomsky—Fodor spectrum.

Next, I survey evidence of the extensive reuse of language circuits across various cognitive domains. This evidence speaks loudest against the conventional wisdom concerning a dedicated faculty of language, and converging evidence from other sources corroborates this view. At the level of implementation, then, it seems language is not special vis-à-vis other cognitive domains. But this then raises the old question about the robustness of language acquisition in children. The evidence of a “poverty of stimulus” continues to baffle many researches across the cognitive sciences, and it is the main motivation behind the persistent and (still) pervasive conviction that language must after all be special. The section following therefore addresses the poverty-of-stimulus issue, but in a spirit rather different from that which has been typical in discussions of linguistic nativism. Instead of throwing mud at the poverty-of-stimulus argument in the hope that some of it sticks (some of it certainly does, but enough people have thrown it for me to feel justified in moving on), I consider how a fairly robust species trait like language can be supported within a thoroughly domain-general architecture. To cap off, I parlay everything canvassed in the discussion up to this point into a general outline of how language could be implemented in the brain so that its autonomy and apparent dissociability may be fully accommodated alongside the evidence of its reuse and relative ontogenetic robustness. Here the Redundancy Model comes to the fore.

As I pointed out earlier, the principle of neural redundancy has not featured prominently in debates concerning the modularity of mind. The basic idea here is that, no doubt for good evolutionary reasons, the brain incorporates a large measure of redundancy of function. We do not seem to exhibit what has been referred to as modular solitarity—a single token module for each type of module that we possess.¹ Instead, we come equipped with very many tokens of the same type of module or brain region densely packed into contiguous regions of cortex. I submit that this fact can account for a lot of what we see when we examine the evidence of cognitive dissociations. More importantly, it can provide an elegant and simple solution to the engineering problem posed by the fact that many of our psychological faculties (speech, problem-solving, playing musical instruments, etc.) seem to require multiple simultaneous use of the same sorts of underlying cognitive mechanisms (the time-sharing problem). There is also evidence that quite often the same sorts of mechanisms are recruited for deliberative “central system” functions on one hand and fast/automatic or “peripheral” functions on the other. This is puzzling because the degree of cognitive impenetrability on display plausibly calls for segregated circuitry (the encapsulation problem). Redundancy naturally explains the data here, too. In fact, for all we know, redundancy might help explain many other quirks of cognition that have so far proved elusive within classical cognitive science paradigms, hostile as they often have been to implementational considerations. The same solution could suffice to solve several problems.

A good chunk of the evidence of reuse comes from neuroimaging data, but as I already indicated in Chapter 3, concerns over the spatial resolution of current imaging technologies have been played as a possible trump card against the idea of the literal reuse of neural circuits. While neuroimaging evidence is not the only evidence on point, and converging biobehavioral evidence also points to the extensive redeployment of the self-same neural technology, still, the likelihood of some cognitive mechanisms running in parallel and in close spatial proximity cannot be discounted, and indeed seems rather high given what we know of the iterative, tessellated, and almost lattice-like arrangement of modules in the cortex. The Redundancy Model beautifully supplements and extends the reuse picture in a way that is completely consistent with the neuroimaging data, faithful to the core principle of reuse, and compatible with the apparent modularization of technical and acquired skills in ontogeny. As I shall explain—and in keeping with the constrained plasticity model I presented in Chapter 6—it chimes with the motto that some modules are “made, not born,” but without crude assumptions about the nearly limitless malleability of cortical tissue. In sum, it gives us just what we need to explain a (mini)modular yet fully domain-general cognitive system within a sensible and neurobiologically informed framework of explanation.