The Adaptable Mind: What Neuroplasticity and Neural Reuse Tell Us about Language and Cognition - John Zerilli 2021
Setting the Scene
A familiar trope of cognitive science, linguistics, and the philosophy of psychology over the past forty or so years has been the idea of the mind as a modular system. In the context of contemporary psychology, a modular system is typically one consisting of functionally specialized subsystems responsible for processing different classes of input, or at any rate for handling specific cognitive tasks like vision, language, logic, music, and so on. The general motivation for this idea is the belief that the mind is complex, composed of different parts that do different things—as opposed to a homogeneous whole that is indivisible and more or less uniform throughout. This “holism,” which modular theories of the mind have been at pains to deny, may not, at first blush, seem like a probable contender for our best explanation of cognition: who would venture to deny that the mind is anything other than complex and intricately organized? But holism isn’t quite the belief that minds are simple. Indeed, traditionally, holism often went along with the belief that minds are impenetrably mysterious and beyond fathoming—something like an ethereal or immaterial substance. What holism does claim, however, is that whatever it takes to make a mind, it isn’t many different things, but many (or a few?) of the same sorts of things. Modularity can be understood as a general rejection of this claim. It avers that in order to make a mind, you’re going to need many different sorts of things, lots of different moving parts.
Though no doubt a plausible, methodologically fruitful, and highly influential idea in its own right—thanks in no small part to Jerry Fodor (1983), whose pioneering effort gave it contemporary theoretical expression and substance—modularity entered the scene in a big way at just about the time that saw the arrival of a new and potentially subversive force in the behavioral and brain sciences: the mature field of neuroscience. Despite its earlier beginnings, neuroscience only really came of age in the late twentieth century, and one of its outstanding achievements has been the discovery of the brain’s lifelong powers of renewal and reorganization. Neuroplasticity has, for better or worse, challenged many of the orthodox conceptions of the mind that originally led cognitive scientists to postulate mental modules. Similarly, rapidly accumulating neuroscientific evidence of the reuse or redeployment of neural circuits, revealing the integrated and interactive structure of brain regions, has upset basic assumptions about the relationship of function to structure upon which modularity—not to say neuroscience itself—originally depended. These movements, developments, and cross-currents are the subject of this book.
Although there are many reasons one might find the “modularity of mind” an interesting hypothesis, an especially compelling one is the possibility that it could account for our linguistic abilities. Indeed, the notion that language is not just phylogenetically special—unique to the human species—but also cognitively special—unique within the overall envelope of our cognitive talents—makes the modularity of mind a wonderfully intriguing hypothesis. Noam Chomsky (1975, p. 4) once remarked that “[t]o come to know a human language would be an extraordinary intellectual achievement for a creature not specifically designed to accomplish this task.” If by “specifically designed” he meant “equipped with a special-purpose faculty of the mind,” it is just this claim that I wish to dissect in this book. Many cognitive scientists continue to believe that language is cognitively special, and hence that its operating principles cannot be assimilated to the rules or procedures engaged within other domains of cognition (such as general problem-solving or pattern recognition). As Annette Karmiloff-Smith (1994, p. 698) once vividly expressed this point: “Lining up objects does not form the basis of word order. Trying to fit one toy inside another has nothing to do with embedded clauses.” The issue is tied up in what is possibly the most contentious and acrimonious dispute in all of modern linguistics: Is language innately specified?
Throughout this book, I shall be concerned predominantly with the following question: Could something having the rough outlines of Jerry Fodor’s module account for language processing? And if not, what sort of module might plausibly take its place, if any? It will be my contention that language is not subserved by a module in Fodor’s sense, at least not in a straightforward way, and this makes that notion misleading as far as language modules go. I shall argue from principles of both neural reuse and neural redundancy that language is facilitated by a composite of modules (or module-like entities), none of which is likely to be linguistically special, and that neuroplasticity provides evidence that few of them ought to be considered developmentally robust, though their development seems to be definitely constrained by features intrinsic to particular regions of cortex.
There is a conspicuous lack of consensus surrounding the status of modules as neuroanatomical entities, in part because modularity has proven itself to be a highly versatile concept, sustaining different research agendas across the biological and mind sciences. Are they functionally dedicated, innately designated (species-constant) regions of “wetware” whose operations may be described by algorithms (Quartz & Sejnowski 1994, p. 726); or are they in the nature of “software” systems having no phylogenetically necessary relation to specific cortical sites, be they dedicated or otherwise? Is there indeed room for both types, or for hybrids combining features of both types (Horst 2011, pp. 224—225, 261—262)? Fodor, perhaps sensing that the real interest of modules lies partly in their functional/neural dedication and ontogenetic robustness, considered that the first description could serve as a paradigm of modularity—a view that has the merit of being in broad agreement with the neurosciences (Bechtel & Mundale 1999; Anderson & Finlay 2014, p. 5; but cf. Doidge 2007, pp. 291—297; Gold & Roskies 2008, p. 354; see §§ 4.2—4.3, this volume, for further detail). Nevertheless, in recent decades, enthusiasts of modularity have been more willing to throw their lot in with alternative proposals or otherwise endorse increasingly anodyne suggestions about what a module really amounts to. Apart from the general explosion of discoveries in the neurosciences, new and dramatic evidence of the precise extent of neuroplasticity and neural reuse has necessitated a shift of emphasis away from implementation. The innateness hypothesis alone looks to be disastrously discredited if the potential for neuroplasticity is indeed as advanced as it appears, since it underscores the crucial role that learning must play in the acquisition of competencies otherwise presumed fixed or defined by characteristic ontogenetic pace and sequencing. The evidence of neural reuse, for its part, indicates that high-level cognitive tasks such as language processing are enabled by highly distributed neural networks composed of very many smaller brain regions or nodes that are themselves multifunctional and domain-general: the selfsame circuits are redeployed over and over again across different tasks and task categories. This discovery potentially undermines the claim that such high-level cognitive feats reflect domain-specific competencies. Quandaries like these have understandably motivated the attempt to rescue the theory through a renewed emphasis on computational design (Jungé & Dennett 2010; Anderson 2010; Anderson & Finlay 2014, p. 5).
Here I shall take as my guiding idealization something closer to Fodor’s paradigm of modularity, the simple reason being that it has by far been the most influential account of faculty psychology in recent decades and the one that overwhelmingly animates, or at least frequently situates, discussions concerning the modularity of language (Chomsky 1980a, pp. 39, 44; 1988, p. 159; 2002, pp. 84—86; Plaut 1995; Pinker & Jackendoff 2005, p. 207; Fitch et al. 2005, p. 182; Collins 2008, p. 155; Fedorenko & Thompson-Schill 2014; see also Karmiloff-Smith 1992). In its neurophysiological and neuroanatomical respects, Fodor’s paradigm module also closely resembles the notion of a brain module familiar to the neurosciences (see Chapter 4). Thus I take a “module” to be something more substantial than a mere “cognitive system.” Specifically, I take a module to be a specialized and autonomous cognitive mechanism that is reliably associated with a unique neural structure, whether local or distributed.1 As far as I can see, only a proposal along these lines has any chance of making modularity interesting and worth pursuing. I take the possibility of functional decomposition—the possibility of understanding a complex subject by breaking it down into different “parts” (regardless of whether these “parts” overlap in the brain)—as being uncontroversial: of course a domain needs to be broken down if we are to have any hope of gaining insights into how it functions! So those insisting that language is modular because it furnishes a fruitful “domain of inquiry” don’t collect any prizes in my book (see § 4.2.3). If I were pressed to stipulate necessary and sufficient conditions, I would lay down functional specificity (i.e., dissociability in principle) as the sine qua non of modularity (Carruthers 2006; Barrett & Kurzban 2006). I explain and defend this position in Chapter 4.
In the interests of full disclosure, let me stress that, by “autonomous,” I do not mean automatic, autonomic, or mandatory; i.e., reflexive (rather than reflective) and therefore independent from central decision and control. I have a somewhat broader notion in mind, with automaticity representing only an especially extreme case. A system in my usage is autonomous when it can even moderately perform without conscious advertence, just as “skills that are practiced over and over acquire a certain degree of autonomy and insularity” (Ohlsson 1994, p. 224). An experienced pianist who does not deliberate over the arpeggios in a well-rehearsed performance, or who holds a conversation as she plays; even a driver holding a conversation as she shifts gears—each capacity displays a measure of autonomy from central control (Rasmussen 1986; Stanton & Salmon 2009; Walker et al. 2015). The operation is still subject to the will, and therefore not quite out of control, but it runs on autopilot all the same.2 Actually, the examples of the pianist and the driver juggling more than one task, with one of the tasks running autonomously, share their juggling-act-like characteristics with language parsing and speech production. Fluent reading, too, where the process of instant character recognition runs autonomously of textual comprehension, is yet another example of information processing sharing features with expert musicianship, fine multitasking motor control, and language parsing (although of course reading, unlike language parsing, requires explicit instruction and drilling). Notice that these observations are consistent with the possibility that at least some modules are “made, not born” (Bates 1999). Often the claim is made that truly interesting modules are innate. But I follow most theorists in regarding innateness as a merely contingent rather than essential feature of modularity (e.g., Coltheart 1999), interesting though it certainly would be if a module were innate (see p. 7).
On the other hand, by “specialized” or “dedicated,” I shall mean more or less what Fodor means when he describes modules as informationally encapsulated, domain-specific, functionally dissociable, and neurally localized. “Informational encapsulation” refers to a module’s restricted access to information outside its own system-specific data store (e.g., a visual module’s being impervious to beliefs the agent has about what she is seeing3), while “domain specificity” refers to a module’s sensitivity to a restricted domain of inputs (e.g., visual, auditory, grammatical, etc.; see the discussion in §§ 2.4.3 and 5.1 for a clarification—and restatement—of this principle). A system is dissociable if it handles a specific function that can be selectively impaired, as when damage to a specific region of the brain causes a selective deficit (e.g., when speech comprehension is impaired, but speech production is not). Finally, a system is localized when it is uniquely subserved by relatively circumscribed or contiguous neural circuitry (Fodor 1983, p. 99; Prinz 2006; Robbins 2009; Gerrans 2014, p. 46). Actually, there is no cogent reason to believe that a module has to be contained within the topographical limits of a specific neural site—its processing parts could well be scattered over the whole cortex, or indeed the whole brain. So the main point of my calling attention to this last feature is to remind the reader that a module has to be exclusively realized somewhere in the brain, even if this “somewhere” is not contiguous.
For convenience, we can refer to a specialized module as an “anatomical” module (Bergeron 2007; Anderson 2010). Occasionally it will be necessary to use the term “specialized” in a somewhat stricter sense than applies to anatomical modules. Specialization in this stricter sense refers to non-reusability across multiple domains, which is essentially a very rigid kind of domain specificity. The clearest examples of units specialized in this stronger sense would be the constituent elements of an anatomical module: the dedication of modular elements to their parent module renders them dedicated or specialized in a strict sense. The parent module will be specialized in a loose sense of the word at least—specialized in the sense that it does some functionally discrete thing, and presumably the same thing every time; but if the parent is reusable across multiple cognitive domains, it will not be specialized in the stricter sense I have in mind (see § 5.1 for an extended discussion).
Now, while the foregoing notion of modules suffices to furnish a general target of inquiry, there are only two features of such modules to which I shall be drawing special attention in this book; namely, functional dissociability and innateness. As I have already foreshadowed, I shall argue that dissociability is the sine qua non of modularity (Chapter 4). This focus at once emphasizes that a module is a special-purpose, special-structure device: to the extent that separately modifiable cognitive systems exist, it is because there are to that extent separately modifiable neural processes. The more the neural hardware is shared (or “reused”) as between distinct cognitive functions, the less likely will such functions be selectively modifiable (barring some principle of neural plasticity or compensation). Presumably, if all the relevant neural structures (circuits, pathways, ions, etc.) were reused across all cognitive functions, cognitive dissociations would not be possible. Such a conclusion seems extreme, but it is implicit even in the view of those who maintain that dissociations are compatible with pervasive neural reuse. Ritchie and Carruthers (2010, p. 289), for example, state that “at the limit, two modules could share all of their processing parts while still remaining dissociable and separately modifiable. For the differences might lie entirely in the patterns of connectivity among the parts, in such a way that those connections could be separately disrupted or improved.” This is plausible,4 and implicit in the conjecture is that some structure (even if it is just the connection pathway between two circuits) must be unique to a function in order for that function to be dissociable at all.
As for the focus on innateness, my defense is that, especially in linguistics, a traditionally very powerful argument for the modularity of linguistic abilities makes much of the fact that language acquisition in children proceeds in a breathtakingly effortless fashion, and in the face of frequently degenerate input. That is, the developmental schedule of language in most normal, healthy children unfolds at such a pace and in such a fashion as to suggest that children are, in Chomsky’s words, “specifically designed to accomplish this task.” I accept that language acquisition in children is impressive, but to make my case in this book, I’ll need to consider whether, even if language acquisition is as effortless and speedy as has been alleged, this developmental phenomenon really does force us back into the arms of modularity. So innateness will join dissociability in my examination of the subject.
Perhaps unsurprisingly, these two features together also typify what seems to be a sort of defeasible starting position in many discussions of modularity (see, e.g., the observations by Barrett & Kurzban 2006, pp. 638, 641; Bergeron 2007)5 and are prominent within the accounts of those defending the existence of a language module.6 In addition, they offer an ideal segue into discussions concerning such functional characteristics as encapsulation and domain specificity (see Fedorenko & Thompson-Schill 2014, and Anderson & Finlay 2014, p. 4; also note Chomsky 1975, pp. 40—41). I’ll explain why this is the case as we go along. At any rate, these are the properties I’ll be evaluating in the light of evidence of neuroplasticity and neural reuse, pursuing the implications of such exciting new discoveries in neuroscience for our understanding of the modularity of mind and language in particular. The aim is to explore candidly what such discoveries suggest about the existence of modules in the robust sense I take to be interesting.
The structure of the book is as follows: Chapters 2 and 3 provide an overview of both neuroplasticity and neural reuse in the human brain. The brain exhibits quite remarkable plasticity. I explore various forms of plasticity, of which synaptic plasticity is perhaps the most important, given its likely role in the formation of cortical maps. This chapter concludes with a brief examination of a special kind of cortical map reorganization; namely, supramodal plasticity. This leads directly on to the notion of neural reuse, which I survey in Chapter 3.
Chapters 4 through 6 pursue the implications of neural reuse and plasticity for the modularity of mind. Chapter 4 presents an overview of the history of modular theorizing about the mind, and uses this historical context to present various conceptions of modularity. I argue against soft conceptions of modularity, and defend dissociability as the sine qua non of modularity. Chapter 5 considers the implications of neural reuse, while Chapter 6 considers the implications of neuroplasticity. Chapter 7 then considers the implications of both neural reuse and neuroplasticity for language. Chapter 8 rounds off the argument by casting doubt on the empirical claim that psychological states are multiply realized. The aim of this chapter is to refute the idea that cognitive science cannot be constrained by neuroscience, an idea that has regrettably obstructed fruitful collaboration between neuroscience and psychology in the past and that could prove to be even more damaging in the future, when evidence of neural reuse looks set to make things a whole lot more interesting. I conclude with some final reflections in Chapter 9.
1 I construe the term “structure” here quite loosely, potentially including ion channels and receptors.
2 Independence from central decision and control is a concomitant of independence from central information (or “cognitive impenetrability”), a special kind of “informational encapsulation” (see next paragraph).
3 This is but one instance of informational encapsulation, often referred to as “cognitive impenetrability.” In general, when I use the term “informational encapsulation,” I mean it in the broadest sense, not in the limited sense denoting cognitive impenetrability alone.
4 But see my § 5.2.
5 Peter Carruthers (2006, p. 2): “In the weakest sense, a module can just be something like: a dissociable functional component.” Admittedly, some, indeed Fodor himself, have nominated other properties such as encapsulation as the sine quibus non of modularity.
6 In most cases, the precise notion of a “language module” at stake is somewhat more fine-grained than this crude description might initially suggest, contemplating such distinct varieties as both Fodor’s sentence parser and the broad language faculty that encompasses Chomsky’s Merge (see my Chapter 7). Notice that in Chomsky’s usage, the terms language “organ,” “faculty,” “module,” and “acquisition device” are used interchangeably (see, e.g., Chomsky 1980a, pp. 39, 44; 1988, p. 159; 2002, pp. 84—86).