The Adaptable Mind: What Neuroplasticity and Neural Reuse Tell Us about Language and Cognition - John Zerilli 2021
What is Neural Reuse?
Neural Reuse and Recycling
Our brief survey of neuroplasticity led us to a consideration of one rather striking feature of neural organization: what is variously termed “supramodal,” “metamodal,” or “amodal” organization. This feature of brain organization makes it possible for a region of the brain typically responsive to a unique stimulus to respond to input mediated by a different modality, and thus enable the cooperation of neural ensembles in the absence of standard inputs. We saw that supramodal plastic changes may be distinguished from crossmodal changes by virtue of the altered regions’ retaining something of their original character and neural function—their contribution has not been wholly, or in many cases even primarily, subordinated to the processing demands of the alternative modality. That such ensembles appear to be operative in normally sighted and hearing adult subjects suggests, furthermore—perhaps somewhat surprisingly—that supramodal organization is a latent feature of the normally functioning brain. We must now stop to consider how such evidence forces us to rethink some basic assumptions in cognitive and neural science. It is not the recruitment of multiple brain areas or modules that gives pause for thought here, for no doubt complex tasks will require a degree of intermodular cooperation. What is striking is the possibility of significantly more overlaps between the neural regions implicated in higher cognitive functions than the standard picture allows, and hence the sharing of neural resources at a much finer level of detail (i.e., in a vastly more promiscuous fashion) than previously acknowledged. Taken overall, the evidence rather suggests that what we might initially think of as basic modular units could resolve into still more basic domain-general (i.e., task-selective) elements, and that hitherto grossly specified functions such as vision and language cannot be located in functionally dedicated regions of the brain. The evidence is thus compatible with the deep interpenetration of higher level psychological functions, as distinct from merely their co-option.
One of the core principles of neuroscience is the principle of functional localization, the idea that specific brain functions “can be mapped to local structure in a relatively straightforward way” (Anderson 2010, p. 245; Gold & Roskies 2008, p. 354). Modern neuroscience is largely predicated on the discovery of such structures and reckons success when a relatively discrete anatomical site can be correlated with some aspect of behavior or function. Still, it has never been entirely clear to what extent, or in just what way, this assumption can be justified. For one thing, some obvious questions immediately obtrude: “The main questions to be answered by any theory that claims that the mind consists of parts are Which parts? and Why those parts?” (Ohlsson 1994, p. 724). Holding that mental functions fall along such axes as language, mathematics, physics, psychology, and so on, calls for a principled defense of this selection, but at times the choice seems a trite folksy, not to say arbitrary. Behind these questions lies the more specific issue of how any supposed carve-up might square with psychological data demonstrating the apparently interactive structure of many behaviors, even those as simple as reflexes (Amaral & Strick 2013, p. 337). How is a fact like supramodal organization, by virtue of which bilateral occipital cortices appear to be standardly redeployed in semantic language tasks, to be accounted for on the assumption that brain areas are highly specialized? At least one thing is abundantly clear: “functional differences . . . cannot be accounted for primarily by differences in which brain regions get utilized—as they are reused across domains” (Anderson 2010, p. 247).
Evidence of the “reuse,” “recycling,” or “redeployment” of brain areas is now extensive (Dehaene 2005; Anderson 2007a, 2007b, 2007c, 2008, 2010, 2014). These terms refer to the “exaptation” of established and relatively fixed neural circuits over the course of evolution or normal development, generally without loss of original function.1 “[R]ather than posit a functional architecture for the brain whereby individual regions are dedicated to large-scale cognitive domains like vision, audition, language and the like, neural reuse theories suggest that low-level neural circuits are used and reused for various purposes in different cognitive and task domains” (Anderson 2010, p. 246). Speaking of an increasingly familiar example of the reuse of an area once thought to be highly specialized, the neurolinguist David Poeppel remarks:
A statement such as “Broca’s area underpins language production” (or “speech,” or “syntax,” or other broad categories of linguistic experience) is not just grossly underspecified, it is ultimately both misleading and incorrect. Broca’s region is not monolithic but instead is comprised of numerous subregions as specified by cytoarchitecture, immunocytochemistry, laminar properties, and so on. And domains of language such as “syntax” are similarly not monolithic but shorthand for complex suites of underlying representations and computations. It is perhaps not surprising that a brain area such as Broca’s region is therefore implicated in many functions, some of which are not even particularly tied to language. . . . Future work ought to focus on “decomposing” or fractionating such complex psychological functions into putative primitive operations to account for the wide range of phenomena that are mediated by anatomically complex brain structures such as Broca’s area. (2015, p. 140)
Language coarsely characterized as a gross function (or subfunction; e.g., recursion) appears to disarticulate into much finer functional granules whose computational resources are available both within and outside the domain of language. This is the essence of the theory of reuse: it explains overlapping neural activation with the suggestion that far smaller functional units with structured operations are used and reused across various task categories. Perhaps many statements that have now attained the status of platitudes—such as “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” (Karmiloff-Smith 1994, p. 698)—have in fact been premature. In what follows here and the next few chapters, I shall certainly argue that this is so, inspired as I am by a commitment to the basic principle that intuitions about cognitive functions need always to be examined (and reexamined) in the light of what neuroscience actually reveals, even where this looks to be at odds with what comparative psychology or linguistics suggests about uniquely human, uniquely linguistic cognitive feats (see, e.g., Chomsky 1965, pp. 58—59). The comparative psychologist might well ask: “If word order is just object discrimination and sequencing, and recursion some sort of applied folk physics, why is it that chimpanzees have nothing even approaching a human language system, though they manifest rich sensorimotor and representational abilities?” There is no shame in confessing that the answer here is by no means clear, which is no doubt why many continue to hold out hope that at the very least some aspects of language processing might not just be uniquely human, but also uniquely linguistic. One dares suggest that there might well be a small or even exiguous component of otherwise highly interpenetrated circuits that is rarely reused outside the language domain, and that would in consequence be specialized in a strict sense—a mechanism recruited for linguistic purposes and little else, dedicated by virtue not only of its isolable functional contribution and circumscribed circuitry, but also its dedication to a specific task category. Consider the possibility of a neuron or tightly restricted set of neurons being dedicated to, say, conjugating the verb “to be” and having no nonlinguistic functions at all (Prinz 2006). This component might aptly be described as a language “module” (or “minimodule”) for all practical purposes (see Chapter 4), and I shall consider its prospects in Chapter 7.
For the present, it suffices to remark that the evidence to which Poeppel refers in the extract cited earlier cannot be ignored either. The fusiform gyrus was rather wistfully hailed as the “face area” after the discovery that it responds to human faces suggested it could be a special-purpose device (Kanwisher et al. 1997). It was later found that the area responds to other categories of objects for which it appears we have expertise, such as cars, birds, and traveling objects (Gauthier et al. 2000). Even the more fundamental notion that ventral visual processing areas are specialized for shape discrimination has been called into question by evidence that information about many objects is distributed across the cortex, and that in some cases their identities can be recovered from low-level activation patterns across several occipital cortices (Haxby et al. 2001; Hanson et al. 2004). I detail further evidence of neural reuse in § 3.3. For the moment, we must turn to consider what is arguably the leading theoretical exposition of reuse attracting serious attention in cognitive science, neuroscience, and philosophy: Michael Anderson’s massive redeployment hypothesis.