NEUROPLASTICITY,

NEURAL REUSE,
AND
THE LANGUAGE MODULE

JOHN ZERILLI

A thesis submitted for the degree of Doctor of Philosophy

of the Australian National University

August, 2017

© Copyright by John Zerilli 2017

All rights reserved
 Note: Figures on pp. 21, 28, 37, 54, 55, 72 and 91
have been removed pending copyright permission

This research was supported by an Australian Government Research Training Program (RTP) Scholarship

2
Except where otherwise indicated, this thesis represents entirely my own work.

_____________________
JOHN ZERILLI
16 August, 2017

Word count: 66,000

3
Man is a machine so complicated that it is impossible at first to form a clear idea of it, and
consequently to describe it. This is why all the investigations the greatest philosophers
have made a priori, that is by wanting to take flight with the wings of the mind, have been
in vain. Only a posteriori, by unraveling the soul as one pulls out the guts of the body, can
one, I do not say discover with clarity what the nature of man is, but rather attain the
highest degree of probability possible on the subject.

La Mettrie, L’Homme machine, 1748

This thesis is dedicated to the memory of David Hodgson AO QC, jurist, philosopher,
and cognitive scientist

4
Acknowledgements

Special thanks are due to Kim Sterelny, Tori McGeer, Richard Menary, Eva Jablonka,
Stephen Stich, Colin Klein, Gualtiero Piccinini, Larry Shapiro, Tom Polger, Matt Spike,
Ron Planer and Liz Irvine.

I’m equally grateful to Don Rothwell and Judy Jones at the ANU College of Law for my
initial introduction to the College and its community of scholars. Without the College’s
regular offers of teaching it is unlikely I’d have been able to sustain my research.
This thesis owes much of its inspiration to the linguist Noam Chomsky. The technical
brilliance, formal beauty and extraordinary precision of transformational-generative
grammar never cease to dazzle me. That he managed to contrive such a system while still
in his twenties is simply astonishing. Anyone who’s prepared a doctoral dissertation in a
technical discipline knows just how brutal the task can be. To subvert then reinvent a
whole technical discipline—in your twenties!—is nonpareil. I’ll be forever grateful that I
was able to meet him personally during his visit to Australia in November 2011. I was
even fortunate enough to be able to discuss with him some of the ideas that have found
their way into this thesis. As it happens, I have arrived at conclusions which diverge from
his in significant respects. But I have not done so lightly.

A second source of inspiration has been the work of the cognitive neuroscientist Michael
L. Anderson, whom I was also privileged to meet, this time at a workshop run by
Macquarie University’s Department of Cognitive Science in June 2016. The idea of
neural reuse had been with me as a kind of premonition for years, indeed from the
moment I first turned away from the practice of law and began to inquire seriously into
matters concerning the mind and its structure. As a 27-year old, having never formally
studied biology, linguistics, mathematics or philosophy, I incautiously submitted a
master’s thesis to the University of Sydney canvassing issues in which some knowledge of
these subjects would have been advantageous (to put it mildly). It was an unmitigated
disaster, and I have ever since wished to eradicate all traces of it, prevented only by the
limits of my jurisdiction over the University’s thesis repository. Curiously enough, I was
actually awarded the degree, albeit on condition that I make a few emendations; but it was
so poorly crafted and misinformed that to this day I can hardly say why. Nonetheless, and
despite my embarrassment, a few ideas in the thesis stood out for being clearly articulated
and not obviously implausible. One was the idea of neural reuse. Of course I didn’t call it
reuse at the time, and had devised a rather clumsy apparatus with which to express my
theoretical inklings. When I encountered Anderson’s own elegantly conceived and much
more skilfully executed theory of “massive redeployment,” I was able to take its
descriptive apparatus on board. Anderson’s influence will be evident to anyone familiar
with his work in the pages which follow.

Beyond professional and intellectual debts, of course, there are those of the heart. Here I
owe a very special debt of gratitude to my partner, Gavin Leuzzi, for enduring with me
the vicissitudes of doctoral study at no small cost. Being a 33-year old Ph.D wasn’t easy.
Living with one would have presented its own set of challenges. He was an unfailing
source of strength and encouragement. I owe him a great deal.

Finally, to my three wonderful friends, Bec Riva, Hezki Symonds and Jesse Hambly, I
say thank you. You were there for me during the darkest times. Bec, you were a godsend.

5
Abstract

What conception of mental architecture can survive the evidence of neuroplasticity and
neural reuse in the human brain? In particular, what sorts of modules are compatible with
this evidence? I aim to show how developmental and adult neuroplasticity, as well as
evidence of pervasive neural reuse, forces us to revise the standard conception of
modularity and spells the end of a hardwired and dedicated language module. I argue
from principles of both neural reuse and neural redundancy that language is facilitated by
a composite of modules (or module-like entities), few if any of which are likely to be
linguistically special, and that neuroplasticity provides evidence that (in key respects and
to an appreciable extent) few if any of them ought to be considered developmentally
robust, though their development does seem to be constrained by features intrinsic to
particular regions of cortex (manifesting as domain-specific predispositions or acquisition
biases). In the course of doing so I articulate a schematically and neurobiologically precise
framework for understanding modules and their supramodular interactions.

6
Table of Contents

Acknowledgements 5
Abstract 6
1 Introduction 9
2 Aspects of neuroplasticity 16
2.1 SCOPE OF CHAPTER 16

2.2 THE NATURE OF PLASTIC CHANGES IN THE BRAIN 16

2.2.1 Definition 16
2.2.2 Synaptic plasticity 18

2.3 NEUROPLASTIC RECOVERY DURING DEVELOPMENT 23

2.4 CORTICAL MAP PLASTICITY 24

2.4.1 Intramodal plasticity 24

2.4.2 Crossmodal plasticity 26
2.4.3 Supramodal (or “metamodal”) organization 28

2.5 SUMMARY 31

3 Neural reuse and recycling 32

3.1 WHAT IS NEURAL REUSE? 32

3.2 THE MASSIVE REDEPLOYMENT HYPOTHESIS 36

3.3 FURTHER EVIDENCE OF NEURAL REUSE 40

3.3.1 Computational modeling 40

3.3.2 Biobehavioural evidence 41
3.3.3 Final thoughts 43

3.4 SUMMARY 44

4 Modules reconsidered: Varieties of modularity 46

4.1 A PIVOT TO THE NEUROSCIENCES 46

4.2 VARIETIES OF MODULARITY 48

4.2.1 Themes and trends 48

4.2.2 Graph theory and network neuroscience 53
4.2.3 Separate modifiability as the touchstone of modularity 56

4.3 THE BRAIN MODULE 67

4.4 SUMMARY 75

5 Modules reconsidered: Whither modularity? 76

5.1 DOES MODULARITY SURVIVE THE EVIDENCE OF NEURAL REUSE? 76

5.2 CAN COMPOSITE SYSTEMS BE DISSOCIABLE? 92

5.3 MODULAR NOTATION 93

5.4 SUMMARY 95

7
6 Are modules innate? 96
6.1 PRELIMINARY REMARKS 96

6.2 IMPLICATIONS OF NEUROPLASTICITY 99

6.3 SUMMARY 104

7 The language module reconsidered 106

7.1 PRELIMINARY REMARKS 106

7.2 DEFINING A LANGUAGE MODULE 109

7.2.1 The meaning of linguistic specialization 109

7.2.2 The domain of language clarified 114
7.2.3 Examples of elementary linguistic units 120

7.3 IS THERE A LANGUAGE MODULE? 123

7.4 IS LANGUAGE INNATE? 132

7.5 ACCOUNTING FOR LINGUISTIC MODULARIZATION 138

7.6 SUMMARY 146

8 Saving faculty psychology: Debunking the argument from multiple realization 148
8.1 NEW DIRECTIONS IN FACULTY PSYCHOLOGY 148

8.2 MULTIPLE REALIZATION REVISITED 152

8.2.1 Preliminary remarks 152

8.2.2 Empirical arguments for multiple realization 152

8.2.2.1 Neuroplasticity 152

8.2.2.2 Convergent evolution 154

8.2.3 A conspectus of recent arguments against multiple realization 155

8.2.3.1 Outline of arguments 155

8.2.3.2 The argument from comparative psychology 155
8.2.3.3 The argument from grains 157
8.2.3.4 The argument from context 158

8.2.4 Conclusion 161

8.3 NEW LOOK FACULTY PSYCHOLOGY 161

8.4 SUMMARY 164

9 Multiple realization and the commensurability of taxonomies: A note on grains 165

9.1 PRELIMINARY REMARKS 165

9.2 RECEPTION OF THE GRAINS ARGUMENT 166

9.3 PROBLEMS WITH THE GRAINS ARGUMENT 167

9.4 SUMMARY 179

10 Conclusion 181
Bibliography 183

8
 1
_____

Introduction

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
(e.g. for vision, hearing, human faces, etc.), or at any rate for handling specific cognitive
tasks. The general motivation for this hypothesis is the belief that the mind is
heterogeneous, in contrast to an earlier view of the mind which posited a uniform
structure. 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 behavioural and brain sciences, the mature field of neuroscience,
which despite earlier beginnings only really came of age in the late twentieth century. One
of the outstanding achievements of neuroscience in the past thirty years 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 which
originally led cognitive scientists to postulate mental faculties. 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 relation of function to structure upon which modularity—not to
say neuroscience itself—originally depended. These movements, developments and
cross-currents form the subject of the present study.

9
 Although there are many reasons one might find the modularity of mind an
interesting hypothesis, from my point of view it is the suggestion that it might account for
language processing that makes it endlessly intriguing. 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.”
Does specialization for language require that language be subserved by dedicated circuits
in the brain? Many, including Fodor, continue passionately to believe so, their faith
undimmed by the passage of years and the inexorable advance of neuroscience. 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 study I shall be concerned with the following question: could
something having the rough outlines of 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), few if any of which are
likely to be linguistically special, and that neuroplasticity provides evidence that (in key
respects and to an appreciable extent) few if any of them ought to be considered
developmentally robust, though their development seems to be definitely constrained by
features intrinsic to particular regions of cortex (manifesting as domain-specific
predispositions or acquisition biases). Along the way I hope to be able to articulate a
schematically and neurobiologically precise framework for understanding modules and
their supramodular interactions.

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

10
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 which 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
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 comprised of very
many smaller brain regions or nodes that are themselves multifunctional and domain-
general: the selfsame circuits are redeployed over and 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, simple reason being that it has by far been the most influential account of
faculty psychology in recent decades and the one which 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

11
than a cognitive system. Specifically, I take a module to be an innately specialized and
autonomous cognitive capacity reliably associated with a unique neural network. From
my point of view, only a proposal along these lines—a special purpose, special structure
device—has any chance of making modularity interesting and worth pursuing. I take the
idea of functional decomposition simpliciter as uncontroversial, and if I were pressed to
stipulate for 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 shall 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 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 is having a conversation as she
plays, even a driver having a conversation as she shifts gears—each capacity displays a
measure of autonomy from central control. The operation is still subject to the will, and
therefore not quite out of control, but runs on autopilot all the same. 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 multi-tasking
motor control, and language parsing (although of course reading, unlike language parsing,
requires explicit instruction and drilling). These observations are consistent with the
possibility that at least some modules are “made, not born” (Bates 1999).1

1 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).

12
 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
seeing2), 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, and localized when it is subserved by
relatively circumscribed or contiguous neural circuitry (Fodor 1983, p. 99; Prinz 2006;
Robbins 2010; Gerrans 2014, p. 46). 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 nonreusability 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 three features of such modules to which I shall be drawing special
attention here, namely innateness, functional dissociability and neural localizability.
These three properties 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)3 and are prominent within the accounts of those defending the

2This is but one instance of informational encapsulation, often referred to as “cognitive impenetrability.”
When I use the term “informational encapsulation,” I mean it in the broadest sense, not in the limited sense
denoting impenetrability alone.
3Peter Carruthers (2006, p. 2) is even more bluntly parsimonious (“In the weakest sense, a module can just
be something like: a dissociable functional component”), though admittedly some, indeed Fodor himself,
have nominated other properties such as encapsulation as the sine qua non of modularity.

13
existence of a language module, 4 presumably because, in the context of language
processing, they offer an ideal segue into more abstract 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). Notice moreover that dissociability and localization are the essential ingredients in
what I have called an anatomical module (Bergeron 2007, pp. 175-176; Anderson 2010, p.
248). I shall hence be evaluating these properties in the light of evidence of
neuroplasticity and neural reuse, pursuing the implications of these exciting new
developments in neuroscience for our understanding of the modularity of mind and the
language module in particular. The aim is to explore candidly what these developments
suggest about the existence of modules in the robust sense I take to be interesting.

The structure of the thesis will be 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 onto the notion of
neural reuse, which is surveyed 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. Chapters 8 and 9
round off the argument by providing a critical reexamination of the doctrine of multiple
realization, casting doubt on the empirical claim that psychological states are multiply
realized. The aim of Chapter 8 is to refute the idea that cognitive science cannot be

4In 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. 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). See Chapter 7.

14
constrained by neuroscience, an idea which has regrettably obstructed fruitful
collaboration between neuroscience and psychology in the past and which 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. Pursuing the matter a little further, Chapter 9
represents something of a concession to those who contend that psychological states are
multiply realized. Insofar as one particular and highly influential argument against
multiple realization is, as I contend, apt to mislead the mind-brain identity theorist, it
behooves me to be upfront about it. But Chapter 9 is in no way intended to detract from
the thrust of the argument of Chapter 8. I conclude with some final reflections in Chapter
10.

15
 2
_____

Aspects of neuroplasticity

2.1 SCOPE OF CHAPTER

This chapter reviews the general science of neuroplasticity with a focus on those aspects
of relevance to the modularity of mind. The need for such a review stems ultimately from
an interest in the implications of neuroplasticity, particularly for the understanding of
early development. While not enough is known about the molecular and cellular
mechanisms underpinning neuroplasticity to warrant definite conclusions about
development, tentative suggestions, grounded firmly in the available evidence, can and
should be put forward. These form the subject of Chapter 6. The present chapter
provides for the most part only a précis of the evidence as it stands.

2.2 THE NATURE OF PLASTIC CHANGES IN THE BRAIN

2.2.1 Definition

Learning raises an interesting question for the cognitive and neural sciences. On the one
hand the nervous system appears to be wired very precisely. On the other hand
mammalian and especially human behaviour can be extremely flexible. If connections
between the main signaling units of the nervous system are set during early development,
how is it that behaviour and its neural underpinnings can be flexible at all? What is the
extent of neural fixity and flexibility in early development and how is it related to the
stability and dynamism exhibited under different conditions in later life (e.g. during

16
learning or rehabilitation)? The best answer so far attempted and empirically
substantiated is the plasticity hypothesis (Kandel et al. 2013, p. 37). This recognizes what
for most of the twentieth century was denied, that even after a critical period in early
childhood the brain retains its plastic potential throughout life. It appears that “chemical
synapses are functionally and anatomically modified through experience and learning as
much as during early development” (Kandel et al. 2013, p. 37). Plasticity is an intrinsic
and persistent property of the nervous system without which it would be impossible to
understand normal psychological function, or indeed pathological and contrapathological
responses to events throughout life (Pascual-Leone et al. 2005, p. 378). Plasticity is not to
be conceived as an occasional or exceptional state of the nervous system—it is in fact its
normal and ongoing condition (Pascual-Leone et al. 2005, p. 379). What is more, similar
mechanisms appear to be at work in both adult plasticity and early development,
suggesting that the mechanisms of adult learning and developmental plasticity are to
some considerable extent conserved (Saitoe & Tully 2001; Kolb et al. 2001, p. 224;
Neville & Bavelier 2001, p. 261). This last point is crucial, since it is really only in virtue
of such parallels that adult neuroplasticity can serve as a window onto early
developmental processes and carry significance for traditional debates in psychology, for
example about the innateness of language. As Laurence and Margolis observe:

Widespread and significant instances of neural plasticity suggests an inherent

openness to the functions that any cortical area can take on. If this is right, then
the brain’s concept acquisition capacities needn’t be innately constrained toward
any particular outcome. Instead cortical circuits might simply form as required to
accommodate a learner’s needs given whatever contingent sensory input has been
received and the wiring that has been previously established. (2015, p. 124)

Neuroplasticity has been defined as “a change (either a strengthening or

weakening) in synaptic efficacy brought about through experience” (Rose & Rankin 2001,
p. 176). In fact synaptic plasticity is only one of a family of brain plasticities falling under
the general banner of neuroplasticity. In its widest sense neuroplasticity refers simply to
“the capacity of the nervous system to modify its organization,” especially in response to
experience, and includes the varied circumstances of normal development and
maturation, learning in both immature and mature organisms, recovery of function after
injury and compensation following sensory deprivation (Neville & Bavelier 2001, p. 261).
At the same time neuroplasticity transverses every level of organization in the brain,

17
synaptic events having counterparts in both higher and lower levels of organization
running all the way from genes right through to complex behaviour (Shaw & McEachern
2001). These facts should not, of course, be taken to suggest that a synaptic definition of
neuroplasticity is necessarily mistaken. Indeed it is just because the synaptic level
continues to provide the best understood and arguably most powerful model of
neuroplasticity available—synaptic plasticity has a probable role in all of the
developmental stages just described, for instance—that it has become customary to regard
synaptic plasticity as broadly representative of the phenomenon. Given my concern with
modules and the likely role of synaptic plasticity in the arrangement and rearrangement of
cortical circuitry (Neville & Bavelier 2001, p. 261; Shaw & McEachern 2001, p. 434),
there is actually good reason for framing the discussion of neuroplasticity here in terms of
synaptic plasticity. Synaptic plasticity supplies a familiar and tractable neurobiological
model for understanding those cases of neuroplasticity which are likely to be of direct
concern to the modularity of mind, namely cortical reorganization and memory
consolidation. Still it is important to appreciate that the term “neuroplasticity” has a
significantly wider scope than the plasticity associated with merely one level of the brain’s
organization; and after a brief treatment of synaptic plasticity revealing the mechanisms
underlying plastic change, I must ultimately turn to consider cortical map
reorganization—an instance of neuroplasticity that ought to be prioritized in any serious
discussion of modularity (Rowland & Moser 2014). (As for the relation between modules
and cortical maps, see the discussion in § 4.3.)

2.2.2 Synaptic plasticity

Neurons are the basic cellular units of the nervous system—self-sufficient, specialized
cells whose primary function is to receive, integrate and transmit information throughout
the body. Any neuron will receive information from potentially many thousands of other
neurons, such connections consisting of microscopically small clefts between the
terminals (“axons”) and receptive fibres (“dendrites”) of adjacent neurons. The clefts
between individual axonal branches and dendritic spines are known as “synapses,” and
synaptic plasticity involves alterations in the strength of such connections. Neural
connections may be strengthened or weakened in a variety of ways, but the most
frequently cited mechanism involves adjustments to the quantity of neurotransmitter

18
released from the presynaptic cell and/or the number of postsynaptic receptors which
determine how effectively the postsynaptic cell can respond to the quantity of
neurotransmitter released presynaptically. Strengthening occurs typically by persistent
stimulation of the postsynaptic cell. A neurotransmitter’s release into the synaptic cleft
initiates a cascade of biochemical events that may lead to the excitation (or “potentiation”)
of the postsynaptic neuron. Research has repeatedly turned up a number of
neurotransmitters, neuromodulators and ions that appear to be crucial for synaptic
plasticity, including glutamate and calcium ions (Ca++). Glutamate is among the most
excitatory of neurotransmitters so far discovered and works by inducing a postsynaptic
calcium influx which, through repeated stimulation, may result in an action potential.
More precisely, the influx of Ca++ leads to increases in the number and efficacy of
postsynaptic α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (or “AMPA”)
receptors, themselves crucial for consolidating synaptic connections by providing the
primary excitatory input drive on the postsynaptic neuron.

Initially, synaptic plasticity was thought to be limited to such molecular

mechanisms alone, entailing few if any changes to the shape of dendritic spines or the
number of axonal branches and sprouts (i.e. neuromorphological changes leading to
“synaptogenesis” and synaptic “pruning”—the establishment of new connections and
elimination of existing connections), while “neurogenesis” (the generation of new
neurons) was understood to be an exclusively developmental process. It is now known
that beyond enhanced signaling between neurons, synaptic plasticity routinely involves
changes to neuromorphology, and that neurogenesis occurs well into adult life, not just
perinatally as was once thought (Rose & Rankin 2001, p. 176; Fuchs & Flügge 2014).

Two varieties of plasticity widely considered to involve changes at the synapse are
cortical map plasticity (otherwise known as representational or topographic map
plasticity) and the cellular changes attendant on learning and memory consolidation
(Buonomano & Merzenich 1998). Cortical map plasticity refers to the detailed remodeling
of cortical maps in response to “behaviourally important experiences throughout life”
(Buonomano & Merzenich 1998, p. 150). Evidenced across different modalities in a
significant number of mammalian species, including humans, cortical map reorganization
results not just from behavioural changes, environment and injury in later life, but is at

19
least partly responsible for some kinds of early perceptual and motor learning
(Buonomano & Merzenich 1998, p. 150). It covers language cross-lateralization
(migration of function from left to right hemisphere) following injury or trauma early in
life (and even later in life, with adequate rehabilitative training), as well as, perhaps most
especially, the plasticity of sensory and motor maps in response to use or trauma. Notice
that in the context of cortical map plasticity it becomes more useful to think of plasticity
as the opening and closing (or broadening and narrowing) of afferent input channels. It is
this plasticity which seems to have most recently captivated philosophers (see further
discussion at § 8.2.2.1).

While the cellular changes involved in learning and memory consolidation are also
thought to depend on synaptic plasticity, it has been far from easy obtaining empirical
confirmation of this connection, or indeed of whether the same plastic mechanisms are
involved in both cortical map plasticity and memory-related synaptic plasticity
(Buonomano & Merzenich 1998, p. 150). Mainstream opinion in the field seems to err on
the side of an affirmative connection on both counts (Buonomano & Merzenich 1998, pp.
152, 153), but Buonomano and Merzenich (1998, p. 179, cf. p. 165) cautiously conclude
that, as for the connection between synaptic plasticity and cortical map plasticity, “we do
not yet have a sufficient understanding of synaptic and cellular plasticity to fully account
for the experimental data on cortical representational reorganization.”

The kind of memory involved is important to clarify here. Of the two broad
classes of memories distinguished by psychologists, i.e. explicit or declarative memory, on
the one hand (comprising both so-called episodic and semantic memory), and implicit or
procedural memory, on the other, it is implicit memory that is usually intended (Rose &
Rankin 2001, p. 176) (see Fig. 1). Termed “procedural” memory on account of its role in
the performance of routine procedures involving neither deliberation nor specific
memories of having carried them out previously (e.g. brushing one’s teeth, tying one’s
shoelaces, riding a bicycle, etc.), it is the memory store for automated action cued by
specific contexts and stimuli. Implicit memory is more likely to reflect the
neurophysiology of learning and memory—respectively the establishment and
maintenance of altered synaptic connections, on one view—than is an explicit form of
memory such as episodic memory; implicit memory epitomizes the rule that “practice

20
makes perfect” and is after all a kind of habituation, which “forms the basis of selective
attention and is thus the foundation of all other forms of learning” (Rose & Rankin 2001,
p. 177). To that extent implicit memory, which is postulated to be represented in vast
interconnections of neural networks across the brain, including the cortex, can be
expected to be of some relevance to modularity, which (from Fodor’s perspective) is also
understood to underlie targeted and automated behaviour and involve interconnected
cortical networks, albeit of a less distributed character.

Figure 1. The different types of memory and their representations in the brain. Source: Kandel &
Siegelbaum 2013, p. 1462.

The clearest case of synaptic plasticity, and one which is likely to play some role
in or otherwise serve as a model for memory consolidation—and possibly many other
varieties of neuroplasticity—is hippocampal long term potentiation (“LTP”), which, as
its name suggests, is the enduring association of neurons through repeated afferent
activation in the hippocampal formation. While its role in learning and memory is not
conclusively established, some such role has been conjectured from its resemblance to
Hebbian plasticity, named after the Canadian psychologist D.O. Hebb. Hebb’s (1949)
influential model of plasticity was advanced to explain the long-lasting changes in
synaptic strength which he hypothesized to underlie learning and memory. He assumed
that stable changes in synaptic efficacy could occur through interactions among neurons:

21
 When an axon of cell A is near enough to excite a cell B and repeatedly or
persistently takes part in firing it, some growth process or metabolic change takes
place in one or both cells such that A’s efficacy, as one of the cells firing B, is
increased. (1949, p. 62)

Hebb’s postulate asserts that simultaneous or rapidly successive pre- and postsynaptic
activity results in a strengthened connection between cells (“cells that fire together wire
together, cells that fire apart wire apart,” as it is often put). This requires a “coincidence
detector” that records the co-concurrent or rapidly successive activity of pre- and
postsynaptic neurons (Buonomano & Merzenich 1998, p. 154). In hippocampal LTP a
subtype of the glutamate receptor, so-called N-methyl-D-aspartate (or “NMDA”)
receptor, serves this coincidence-detecting function by facilitating the postsynaptic influx
of Ca++ (which, if persistent, typically results in a strengthened connection via increased
AMPA receptor efficacy, as we saw earlier). Since LTP appears to reflect something like
Hebbian associative plasticity, many neuroscientists have not hesitated in postulating
LTP as the neurochemical basis of learning and memory. It has the unique
phenomenology, induction characteristics and longevity “to place it firmly as a candidate
for the storage of experiential memory” (Teyler 2001, p. 101).1

While LTP is generally regarded as crucial to memory storage, some

neuroscientists are more circumspect, either denying that the evidence of LTP
subserving learning and memory is strong enough to justify the faith placed in this
mechanism (Cain 2001, p. 126), or holding out that LTP might instead be “a generic
mechanism for increasing synaptic gain throughout the brain whenever increases in
synaptic strength are needed,” and therefore “a general purpose mechanism by which
synapses can increase their influence…regardless of the kind of circuit in which they are
embedded” (Teyler 2001, p. 105). An equally pessimistic estimate has it that “if LTP
occurs naturally in the behaving animal, it can at best be said to underlie circuit
formation, not learning or memory” (Shaw & McEachern 2001, p. 434). LTP may then,
on a minimal reading, be simply a means by which neural networks are formed and
maintained. But what few would deny is that LTP is an important neurophysiological
substrate supporting various manifestations of neuroplasticity. In fact if the connection

1There is evidence that cortical LTP shares many of its properties with hippocampal LTP (Buonomano &
Merzenich 1998, pp. 157, 174).

22
between implicit memory and modules is rightly drawn, LTP, even on a deflationary
view, could be seen as offering support to the idea that similar synaptic mechanisms are
implicated in the consolidation of memory, the development of modules and the
migration of cortical maps, which like memories and modules are also represented in
stable, if more local, networks of neurons in the brain. (Rowland & Moser (2014) present
evidence that even episodic memory has modular organization, resembling the
neuroanatomical and neurophysiological features of sensory and motor cortical maps, e.g.
columnar structure and topographic arrangement. See §§ 2.4 and 4.3 for elaboration.)

2.3 NEUROPLASTIC RECOVERY DURING DEVELOPMENT

While critical period plasticity may under the right set of circumstances be “reopened” in
later life, the potential for plastic recovery following injury is still very much a function of
age. During development, so-called spontaneous changes to the brain resulting from
injury are likely overall to reduce the plasticity of the region affected; “[i]n contrast, when
the brain fails to change in response to injury there is considerable capacity for
modification of cortical circuitry,” particularly through experience, and here the general
rule seems to be that the earlier the injury and therapeutic intervention the better the
chance of functional recovery (Kolb et al. 2001, pp. 236-237, 239). This is ostensibly
because earlier interventions influence spontaneous changes “in such a way as to
maximize functional recovery.” On the whole, younger animals are more plastic than
older ones, both when it comes to experience- or activity-dependent learning and
spontaneous recovery from injury (Shaw & McEachern 2001, p. 430).

The developing brain is obviously different at different stages, so the character of

spontaneous responses to injury can naturally be expected to differ with age. (Whether
these responses are beneficial will also depend on age.) The best studied case of
mammalian plasticity is probably in the rat. Neurogenesis in the rat is essentially
complete by birth, and produces a cortex that is initially equipotential. Between 7-10 days
of age, the process of cell migration in the cortex—a process which begins well before
birth—comes to an end, at which point activity-independent cell differentiation begins.
This process itself ends by about 15 days of age (i.e. at about the time of eye opening),
although synaptogenesis continues for a further 2-3 weeks beyond this point.

23
Compensation for injury suffered during neurogenesis can be quite extensive (Kolb et al.
2001, pp. 228-229). Even the killing of all cerebral neurons by X-radiation appears to
provoke regeneration resulting in up to 50 per cent of the cerebrum being rebuilt (Kolb et
al. 2001, p. 229). Injuries occurring during the period of cell migration and
differentiation, however, are functionally devastating, with effects even more pronounced
than those caused by the same injuries in adult rats. Then again during the period
immediately following this—and therefore concurrent with a period of intense
synaptogenesis—the brain’s capacity for recovery seems to be optimal (Kolb et al. 2001,
p. 230).

Just why young neurons are more plastic than older ones is unclear, but a very
plausible hypothesis attributes it to the impact of homeostatic mechanisms after the
critical period (Shaw & McEachern 2001, pp. 443-444). The absence of homeostatic
regulatory mechanisms (like lateral inhibition) during critical periods means that
potentiation is ubiquitous and the central nervous system highly unstable. Later,
however:

homeostatic regulation of receptors and synapses becomes paramount, and lateral

inhibition becomes a dominant feature of neural circuits and the interaction
between systems. Given such mechanisms of global homeostasis, the alterations
that do occur in the adult CNS only do so in response to the strongest stressors.
(Shaw & McEachern 2001, pp. 443-444)

One upshot of this explanation is that in one sense the brain remains intrinsically as
plastic as ever, its plastic potential merely suppressed by mechanisms that can themselves
be reversed, as we now know they can, “under precisely defined and controlled
conditions.”

2.4 CORTICAL MAP PLASTICITY

2.4.1 Intramodal plasticity

The most convincing evidence of cortical map plasticity comes from studies of plastic
changes to adult primary sensory cortices. Sensory cortical areas relating to touch, vision
and hearing “all represent their respective epithelial surfaces in a topographic manner”

24
(Buonomano & Merzenich 1998, p. 152). This means neighbouring cortical areas respond
to neighbouring sensory receptors. Somatosensory cortex maps areas of the skin’s surface
somatotopically such that “neighbouring cortical regions respond to neighbouring skin
sites.” Likewise auditory cortices map tones tonotopically, and visual cortices map
features of the visual field retinotopically. Close to three decades of research now confirm
the potential for these sensory cortices and their somatotopic, tonotopic and retinotopic
coordinates to undergo plastic changes in a use-dependent manner (Buonomano &
Merzenich 1998, p. 152).

The plastic changes in view here could well include the recovery of function after
injury to the cortex, for example language cross-lateralization following trauma (Polger
2009, p. 464; Clark 2009, p. 365). In such cases a certain psychological function, be it
tactile, visual, auditory, motor or linguistic, is mediated by a specific region of cortex at
time t1, and by a different region of cortex at time t2 (Polger 2009, p. 464). A particularly
striking example of this is seen in the case of children who develop normal or near-normal
language abilities after a left hemispherectomy, in which the left cerebral hemisphere
(which typically mediates language) is either disabled or removed in its entirety
(Laurence & Margolis 2015, p. 123). A child known as EB was found to have recovered
most of his language skills two years after undergoing a left hemispherectomy at the age
of two and a half and tested as virtually normal with respect to linguistic ability at age
fourteen, his language faculty now subserved by regions in his right hemisphere (Danelli
et al. 2013).

While such instances of plasticity are certainly impressive, and reveal that the
phenomenon is not confined to sensory-motor cortices alone, more typical examples
(indeed the first to be discovered) involve the expansion of cortical maps to neighbouring
regions of intact cortex that have been deprived of sensory input from within the same
modality as that subserved by the invading cortex, for example as might occur when the
cortical area corresponding to one manual digit invades the neighbouring area
corresponding to the adjacent digit following a loss of input to the adjacent digit
(Rauschecker 2001). This phenomenon is known as “intramodal” plasticity. The earliest
studies of neuroplasticity reported intramodal effects in adult monkeys. Using the
topographically arranged somatosensory cortical map as the dependent variable, it was

25
found that when deprived of input, either by median nerve transection or digital
amputation, though initially unresponsive it did not remain unresponsive and was soon
activated in response to adjacent inputs (22 days in the case of transection, 2-8 months in
the case of amputation). Similar results were reported after denervation or amputation in
the raccoon, flying fox, cat and rat, and “large-scale remodeling can occur in human
somatosensory and motor cortical areas in the weeks or months immediately following
limb amputation” (Buonomano & Merzenich 1998, pp. 163, 165). The results are equally
dramatic for the visual and auditory cortices, demonstrating that “when a given cortical
area is deprived of its normal afferent inputs, it reorganizes so that the deprived area
becomes responsive to sensory inputs formerly represented only within the cortical
sectors surrounding those representing lesioned input sources” (Buonomano &
Merzenich 1998, p. 167).

It is as well to note that intramodal plastic changes may be induced without

sensory deprivation. Studies on somatosensory, visual and auditory cortices show that
intramodal plastic changes can occur by training animals on specialized tasks. In humans,
magnetoencephalography (MEG) reveals that hand representations of Braille readers are
significantly larger for the right index finger than for the left index finger or for the right
index finger of non-Braille readers (Pascual-Leone & Torres 1993). Likewise the digital
representation of string players is larger for the left hand than for the right hand or the
left hand of control subjects (Elbert et al. 1995).

2.4.2 Crossmodal plasticity

Whereas intramodal plasticity (as its name suggests) occurs within a modality,
“crossmodal” reorganization involves “expansion of maps in one modality as a result of
deprivation in another” (Rauschecker 2001, p. 244). The changes here are more obviously
compensatory. Cortical maps used for, say, hearing, might project into occipital cortex
following deprivation of visual stimuli, whereupon occipital cortex acquires the
processing structures typical of auditory cortex; or visual deprivation might lead to
recruitment of primary visual cortex for tactile processing (Noppeney 2007). And since
the area supporting the lost function is put to an alternative use, crossmodal plasticity
actually makes recovery of original function quite challenging (Pascual-Leone et al. 2005,

26
p. 395). While it had previously been supposed that interventions must be drastic to
induce crossmodal plastic changes, “it is now clear that simply withholding the normal
pattern of sensory experience in one modality is sufficient to reorganize the neural
representation of the remaining senses”; furthermore, “[i]t appears that the same synaptic
mechanisms are invoked that also rule synaptic changes within the same modality”
(Rauschecker 2001, pp. 244-245). Crossmodal changes require that cortical maps receive
input connections, albeit indirectly, from new epithelial surfaces, and there are essentially
two ways for this to occur: either via synaptogenesis, in which new connections are
established between the deprived cortical region and a region which already has the
relevant connections to the sensory end-organ; or the “unmasking”
(strengthening/rearrangement/potentiation) via LTP or some other synaptic plastic
mechanism of existing connections between the deprived cortex and the sensory end-
organ and/or its associated cortex (Rauschecker 2001, p. 255; Ptito, Kupers et al. 2012).
Unmasking is likely preliminary to synaptogenesis (Pascual-Leone et al. 2005, pp. 379,
394-395; Merabet & Pascual-Leone 2010, p. 48). There is experimental support for both
mechanisms in crossmodal plasticity, and both are likely to play a role in intramodal
plasticity.

The extent of crossmodal plastic change is of course partly a function of time

(Noppeney 2007). Short term changes that enhance the processing capabilities of spared
modalities are probably the effects of unmasking, and consequently more readily
reversible after input restoration (Pascual-Leone et al. 2005, pp. 390-391; Noppeney
2007, p. 1177). Blindfolding induces rapid changes that are just as swiftly reversed after
visual input restoration. Long term deprivation, on the other hand, is more likely to result
in sustained structural reorganization through synaptogenesis following initial unmasking
(Pascual-Leone et al. 2005, pp. 390-391). This would no doubt explain why the most
dramatic crossmodal impacts are observed in cases of early onset and congenital
blindness: “functional reorganization is particularly pronounced in early onset blindness”
(Noppeney 2007, p. 1170). The occipital cortices of such subjects, for instance, appear to
be functionally important for Braille character identification (although not detection),
suggesting a functional contribution of the reorganized occipital cortices in complex
tactile discrimination (Noppeney 2007, pp. 1173-1174). Early and congenitally blind
subjects routinely outperform sighted subjects in both episodic and semantic memory

27
tasks and may even require the occipital pole for higher-level cognitive and semantic
processing (Noppeney 2007, pp. 1171, 1174).

2.4.3 Supramodal (or “metamodal”) organization

Not only congenitally and early blind subjects but sighted subjects too have been found to
exhibit occipital cortex activation during nonvisual information processing (Leo et al.
2012, p. 2). The activation in such cases, however, is not straightforwardly crossmodal,
since it requires neither sensory deprivation nor special training. While any activation of
occipital cortices in sighted subjects performing nonvisual tasks might be ascribed to a
preference for visualizing nonvisual afferents, the same response pattern in congenitally
blind subjects—by definition lacking vision since birth—reveals that some other principle
of cortical functional organization is involved. In these cases occipital cortices do not
merely serve as the site for nonvisual information processing, as might be presumed to
occur in a standard case of crossmodal plasticity, but seem to be contributing something
visual to the nonvisual input, and this is no less true for blind subjects (Striem-Amit &
Amedi 2014, see below). That is to say nonvisual information is apparently being
processed visually, in contrast to crossmodal plasticity which would (presumably) involve
the nonvisual processing of nonvisual afferents, albeit in primary visual cortex. Variously
termed “supramodal,” “metamodal” or “amodal” organization (Pascual-Leone &
Hamilton 2001; Striem-Amit & Amedi 2014; Laurence & Margolis 2015), evidence for
the phenomenon came originally from studies of the dorsal and ventral visual pathways,
implicated, respectively, in space and motion discrimination and object/shape category
recognition (the “where” and “what” visual streams, see Fig. 2). More recently,
supramodally active regions have been confirmed beyond the occipital cortices (Leo et al.
2012, p. 2).

Figure 2. The “where” (top) and “what” (bottom) visual processing streams.

28
 The nature of supramodal organization is best illustrated by studies involving
early and congenitally blind subjects. The dorsal (“where”) visual pathway of such
subjects is active during tactile and auditory motion discrimination tasks and reflects the
activation patterns of sighted controls performing corresponding tasks (Ptito, Matteau et
al. 2012, p. 2). Similarly the ventral (“what”) visual pathway of early and congenitally
blind subjects is active during both haptic (tactile) and nonhaptic (electrotactile) object
exploration tasks, again reflecting activations observed in sighted controls performing
corresponding tasks (although blind subjects activated larger portions of the ventral
stream during nonhaptic tactile shape discrimination than sighted controls) (Ptito,
Matteau et al. 2012, p. 2). In a very recent study it was shown that visual experience in
the perception of body shapes is not necessary for the activation of the visual extrastriate
body area (EBA) (Striem-Amit & Amedi 2014). Congenitally blind subjects were trained
to use a “visual-to-auditory sensory substitution device” which converts visual images
into auditory “soundscapes.” The EBA was robustly active when subjects were presented
with body soundscapes. Hence “despite the vast plasticity of the cortex to process other
sensory inputs” (i.e. crossmodal plasticity), these findings suggest “retention of functional
specialization in this same region” (Striem-Amit & Amedi 2014, p. 4). The dorsal and
ventral processing streams, and the EBA in particular, appear to be modular,
developmentally constrained and functionally preserved despite complete early and
congenital visual impairment. That they are responsive to sensory information channeled
from other modalities also suggests that these regions are not strictly domain specific,
since they are not beholden to specific sensory transduction pathways. Instead they seem
to be sensory-independent and task-selective (Striem-Amit & Amedi 2014, p. 5). The
preexisting intermodal connections that are unmasked under crossmodal influence may,
apparently even in the absence of crossmodal plastic unmasking, supply the critical
cortical infrastructure supporting this supramodal dynamic (Pascual-Leone & Hamilton
2001, p. 439; Pascual-Leone et al. 2005, pp. 393-394; Leo et al. 2012, p. 2). The original
motivation for domain specificity might have been rationalized in roughly the following
way. Any module must (minimally) have a specific function which it “knows” to perform
on just the right occasion/s. Cognitive scientists can explain this with the suggestion that
a specific input or external stimulus cues the module to respond (Pascual-Leone &
Hamilton 2001, p. 431). What supramodal organization vividly demonstrates, however, is
that inputs need not be external stimuli—internally mediated stimuli across modalities

29
are normal—and that any one module will typically be sensitive to more than one
stimulus, including those channeled along intermodal pathways. Put another way, it
would appear that modules are frequently reused.2 I shall explain this in greater detail in
the following chapter.

Compelling evidence of supramodal organization also comes from subjects whose

senses are intact. (This material does not speak to the kind of plasticity we have been
considering so far in this chapter, but it is related in ways that will be clearer in the next
chapter, as well as Chapter 6.) It had already been known that unisensory cortices may be
active when presented with stimuli coming through other modalities, as when a single
stimulus component of a typically bimodal event with a close semantic connection is
presented on its own, for example the sound of tools, the voice of a loved one, the sight of
lips mouthing words, and such like (Hirst et al. 2012). Learning and conditioning of
arbitrary pairings of unrelated stimuli may also produce these results (Hirst et al. 2012, p.
2). What was not confirmed until recently is whether these results depended on a prior
semantic association, or otherwise “an explicit conditioning paradigm, or prolonged,
habitual co-occurrence of bimodal stimuli” (Hirst et al. 2012, p. 2). Hirst et al.’s (2012)
clinical study confirmed that even without sensory deficits, training or semantic
associations primary visual cortex exhibits an increased number of active neurons when
presented with sounds alone provided subjects are preexposed to the auditory and visual
stimuli. There is also evidence that the occipital cortex of sighted subjects is active during
tactile processing of orientation and, perhaps most astonishingly, that semantic word
generation in sighted subjects depends partly on bilateral occipital cortices, regions that
have always been supposed to be among the most specialized in the brain (Pascual-Leone
et al. 2005, p. 394). Studies by Antonio Damasio and Alex Martin were among the first to
demonstrate activation of motor areas during verb retrieval tasks and visual areas during
noun processing tasks such as naming colours and animals (Damasio & Tranel 1993;
Damasio et al. 1996; Martin et al. 1995; 1996; 2000). Merely the sight of manipulable
artifacts, indeed just seeing their names, activates parts of the brain associated with
prehension (Chao & Martin 2000).

2 Cf. Barrett & Kurzban (2006, pp. 634-635), who argue that something like task selectivity defines formal
domain specificity, although it is often enough construed as evidence of a domain-general system: they
observe that “there is no natural line that separates domain-specific from domain-general mechanisms.” See
§ 5.1 for elaboration.

30
 The material presented in this chapter is by no means intended to serve as an
exhaustive or even necessarily comprehensive account of the fascinating field of
neuroplasticity. But what I have provided ought to be sufficient to support the claims I
make in Chapter 6. In the next chapter I provide a synopsis of what could well be
regarded as yet another class of neuroplastic responses, responses which are, however,
sufficiently distinctive in character when compared with cortical map plasticity and
memory consolidation as to warrant separate consideration.

2.5 SUMMARY

The brain exhibits an impressive degree of plasticity. Plasticity is really an intrinsic

feature of the nervous system, not an exceptional or occasional state. Neuroplasticity
comprises a family of different types of plasticity. Of these synaptic plasticity is perhaps
the best understood variety and plays an important role in cortical map reorganization
and memory consolidation. Cortical map plasticity is of direct relevance to any discussion
of modularity. There are two types of cortical map plasticity: intramodal and crossmodal.
Crossmodal plasticity is likely to arise from the underlying supramodal (or “metamodal”)
organization of the brain.

31
 3
_____

Neural reuse and recycling

3.1 WHAT IS NEURAL REUSE?

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 for the cooperation of neural ensembles in the absence of
standard inputs. We saw that supramodal plastic changes may be distinguished from
crossmodal changes in 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

32
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 behaviour 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 defence 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 behaviours, even those as
simple as reflexes (Amaral & Strick 2013, p. 337). How is a fact like supramodal
organization, in 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. “[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

33
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. For
example, in addition to language-specific functions such as syntactic processing or
phonology…functional imaging studies have attributed to Broca’s area the
processing of hierarchically organized motor actions as well as rhythm processing.
While such functions are related to language processing in a broad sense, they
apply to many other domains of cognition. 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
motherhood status—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

34
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 which would in
consequence be specialized in a strict sense—a mechanism recruited for linguistic
purposes and little else, dedicated in 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
might 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.

35
 3.2 THE MASSIVE REDEPLOYMENT HYPOTHESIS

Neural reuse theories comprise what Anderson describes as “an emerging class of
theories” which “taken together…offer a new research-guiding idealization of brain
organization” (Anderson 2010, p. 246). Anderson’s own hypothesis builds on the
assumption that evolution might prefer the reuse of neural circuitry over the development
of new circuitry de novo (Anderson 2010, p. 246). On this assumption three predictions
are thought to follow, the most obvious being neural reuse itself. “A typical brain region
[should] support numerous cognitive functions in diverse task categories.” Second, older
brain areas should ceteris paribus be reused more than newer ones, because “having been
available for reuse for longer” they are more likely candidates for integration into recently
evolved functions.1 Third, recently evolved functions should be more distributed than
older ones since it should on the whole prove easier to utilize available circuits than to
devise special purpose circuitry afresh, “and there is little reason to suppose that the
useful elements will happen to reside in neighboring brain regions.” Conversely, “a more
localist account of the evolution of the brain would…expect the continual development of
new, largely dedicated neural circuits” for every cognitive innovation or significant
increase in cognitive power.

Anderson has tested these predictions in a number of studies with conspicuous

success (2007a; 2007c; 2008). For instance, the typical cortical region was found to be
implicated in fully nine domains extending from action, vision and audition through
language, mathematics, memory and reasoning. This illustrates an important feature of
reuse, i.e. the possibility (in principle) of congruously overlapping regions—just the same
circuits exapted for one purpose can be exapted for another provided sufficient
intercircuit pathways exist to allow alternative arrangements of them. The same parts put
together in different ways will yield different functional outcomes, just as “if one puts
together the same parts in the same way one will get the same functional outcomes”
(Anderson 2010, p. 247, my emphasis) (see Fig. 1).

1 The evolutionary psychologist’s invocation of so-called “debugging” concerns are addressed in § 5.1.

36
 Figure 1. Two cognitive functions
indicated by solid and dashed
lines, organized in the top figure
the way that an anatomical account
of modularity would predict, and
in the bottom figure in accordance
with how neural reuse sees the
matter. Anatomical modularity
maintains functional dissociability
and localization for gross or high-
level functions with few if any
overlapping units. Reuse suggests
overlapping units that form
different patterns of connection.
Source: Anderson 2010, p. 247.

Regarding the second prediction that older areas are more likely to be reused than
recently evolved regions, if we make the simplifying assumption that older areas lie at the
back of the brain, Anderson’s results confirm the expectation. Anderson reports a
negative correlation between the position of a brain region along the Y-axis in Tailarach
space and the number of tasks that activate the region. (Tailarach space is the three-
dimensional human brain atlas used by neuroscientists for mapping locations in brain
space—the correlation is counterintuitively reported as “negative” because in Tailarach
space the origin is set at the centre of the brain with regions posterior measured in
negative coordinates.) The results were replicated using different data sets, Anderson
evaluating them in this vein:

Although the amount of variance explained in these cases is not especially high,
the findings are nevertheless striking, at least in part because a more traditional
theory of functional topography would predict the opposite relation, if there were
any relation at all. According to traditional theories, older areas—especially those
visual areas at the back of the brain—are expected to be the most domain
dedicated. But that is not what the results show. (2010, p. 247)

As for the third prediction, that recently evolved functions ought to generate
more distributed patterns of activation than older ones, Anderson’s (2007a) findings
suggest that language could well be the paradigm, supported by more distributed
activations than visual perception and attention and indeed any other domain that was
tested (Anderson 2008). Results such as these showing widely scattered activations across

37
the brain for putatively late-developing functions are incidentally consistent with the
degree of specialization for local circuits that neural reuse actually presupposes. For
neural reuse is of course consistent with a certain kind of specialization; as the very word
“redeployment” suggests, it presupposes the existence of comparatively fixed neural
circuits whose functional contribution may be preserved across multiple task domains.
The metabolic costs of maintaining long-distance connections would presumably
encourage the reuse of local flexible (“poly-functional”) circuits, if any were around;
“[t]hat this is not the observed pattern suggests that some functionally relevant aspect of
local circuits is relatively fixed” (Anderson 2010, p. 247, but cf. Anderson 2014, pp. 15-
16, 104).2 Anderson’s earliest formulations of the redeployment hypothesis accounted for
this fixity by introducing an important distinction, following Bergeron (2007; 2008),
between stable low-level computational “workings” (or cortical “biases”) and diverse
high-level cognitive “uses.” Workings are represented in the numbered units of Figure 1,
above, while uses are represented by the functional composites formed from these units.
Workings are really very tiny regions of cortex that make a specific computational
contribution to higher-level cognitive tasks or “uses.” We might say that workings
represent an essential functional contribution across all task categories, considered in
isolation of neural context (although Anderson has since moved away from essentialism),
and that uses are the high-level cognitive functions enabled by the composite of several
workings. (More on this in Chapter 4.)

Stable structure for local circuits is a feature of a closely related account of neural
reuse, one which posits reuse or recycling as a developmental solution (in contrast to
Anderson, for whom reuse is primarily an evolutionary solution). How are readily
transmissible cultural practices whose phylogenetic emergence is too recent for
evolutionary hardwiring to explain, such as reading and arithmetic, to be
neurophysiologically accounted for? Early developmental neuroplasticity might be one
way, but in supposing that local circuits might be too rigid for the effects of experience to
overcome, Dehaene (2005) gives priority to “neuronal recycling.”

2Scattered activations can be explained other than by the suggestion that local circuits are relatively fixed. I
return to this issue in Chapter 5 (see § 5.1).

38
 Of course neuroplasticity is not something either Dehaene nor Anderson would
wish to deny. Dehaene goes as far as accepting (as I think one must) that novel uses which
depart significantly from existing cortical biases cannot simply be exapted from them: a
high-level use which places a significant cognitive burden on existing circuits, themselves
better suited for other uses, must in the end disrupt those circuits and the alternative uses
to which they might be put. More cognitively demanding functional acquisitions
therefore require more neuroplasticity. This brings us to a potentially thorny issue. Just
what is the relationship between reuse and plasticity? There might seem to be a tension
between the fixity necessary to get reuse off the ground, on the one hand, and the
plasticity necessary for reuse to play an interesting role in learning and evolutionary
novelty, on the other. In fact there is no real problem here (Anderson & Finlay 2014).
Anderson describes neural reuse as a change in use without a change in working, and
plasticity as a change in use resulting from a change in working (Anderson 2010, p. 297).
There is no real problem here because some forms of neuroplasticity (such as Hebbian
synaptic plasticity) do not require flexible units before they can effect a change in use,
given that they involve only adjustments to connection strength (“a change in use without
a change in working”); besides that, neural units are not quite as “fixed” as Anderson’s
own (2010) remarks suggest, allowing for more drastic forms of neuroplasticity (such as
synaptogenesis and the like) to partly override the natural biases of particular brain
regions (“a change in use from a change in working”). I revisit these themes in Chapter 6.

It might be just as well to point out one other respect in which the story of neural
reuse is compatible with the known biophysical constraints and possibilities of neural
circuits. Neural reuse is really an ideal solution to what might be called the scaling
problem (Zador 2015, p. 43; Bullmore & Sporns 2012, pp. 337-339). The scaling problem
refers to the dilemma that as the number of neurons increases, undoubtedly conferring
advantages in the form of increased behavioural flexibility and intelligence, the number of
neurons that must be connected before such advantages can materialize grows
quadratically larger. Thus in a small 10 neuron circuit, only 100 connections are required,
but in a larger circuit consisting of perhaps 100 million neurons anywhere up to a
thousand billion connections might be required. It is not easy to see, from the point of
view of engineering and design, how an ever larger brain can be wired up efficiently when
the number of neural connections required soon becomes immense. This consideration

39
has actually been played as an argument in favour of modularity, but it could just as well
be pressed into the service of neural reuse, which delivers fixed low-level cognitive
workings that operate autonomously in something like the way of traditional modules (see
§ 5.1).

3.3 FURTHER EVIDENCE OF NEURAL REUSE

3.3.1 Computational modeling

A number of large-scale computational models of the brain are currently being developed
in the hope of understanding the activity of a million neurons or more. At their most
sophisticated they leave behind the biologically unrealistic neural network models of the
past and enter the domain of real brain simulation, neurorobotics and neuromorphic
computing. As the computational analogues of real neural networks, they are beginning to
offer fresh insights into the brain’s dynamic response properties. The primary advantage
of brain simulation is that, “unlike the empirical brain, the model’s internal workings are
completely known and the model’s structure can be modified in order to explore how its
activity changes” (Sporns 2015, p. 97). One example of brain simulation that is especially
relevant in the present context is Spaun (Semantic Pointer Architecture Unified
Network). Spaun has a single eye through which it receives digital images as input, and a
moving arm through which it provides behavioural output (Eliasmith 2015). What is
interesting is that its 2.5 million neurons are organized to simulate about twenty of the
roughly one thousand functionally distinct areas which neuroscientists typically identify
in the brain (e.g. separate neurons for frontal cortex, basal ganglia, occipital cortex, etc.).
One feature of Spaun that supports the theory of reuse comes as a result of this unique
“modular” architecture:

One key contribution of Spaun relative to many competing architectures is that

Spaun can perform a variety of different behaviours, much like an actual brain.
For example, Spaun can use its visual system to recognize numbers that it then
organizes into a list and then stores in working memory. It can then later recall
this list and draw the numbers, in order, using its arm. Furthermore, Spaun can
use this same visual system to parse more complex input…To do so, it uses the
same memory system, but in a slightly different way. As well, it uses other brain
areas that it didn’t use in the list recall task. That is, Spaun can deploy the same

40
 brain areas in different ways depending on what task it needs to perform.
(Eliasmith 2015, p. 132, my emphasis)

Spaun’s differentiated circuits manifest the very same interactive dynamics which reuse
posits of real brains: “different, specialized brain areas are coordinated in a task-specific—
that is, flexible—way to meet a challenge presented by the environment” (Eliasmith 2015,
p. 132). This behavioural flexibility marks a distinctive sense in which neural reuse is a
form of plasticity: the ability to switch effortlessly from task to task (reading an email,
playing a piano, chasing a dog) using the same brain areas in different ways and with little
or no delay in shuffling between them. This kind of plasticity serves to set biological
intelligence apart from most contemporary artificial intelligence, and indeed explains why
Spaun is “atypical of the field” overall (Eliasmith 2015, p. 134). Most machines are good
at doing one specific thing (playing chess, solving mathematical equations, driving a car,
etc.). Spaun is unique both in the variety of tasks it can perform and its capacity to learn
new behaviours independently “while preserving abilities it already has” (Eliasmith 2015,
p. 134). Spaun may be one of the first tentative steps towards showing that a domain-
general learning system can work.3

3.3.2 Biobehavioural evidence

Casasanto and Dijkstra (2010) report an interesting association between autobiographical

memory and motor control. The task involved shifting marbles upward or downward
from one container to another while relating memories having either positive or negative
valence. Subjects were asked to retell for example a negative memory, followed by
another negative memory, then perhaps a positive memory, while simultaneously moving
marbles from one container to another in a given direction. It was found that subjects
retrieved more memories, and moved the marbles more quickly, when the direction of
movement aligned with the valance of the memory, i.e. when the upward movement
coincided with positive memories, and the downward movement with negative memories.
Even when subjects were not asked to relate memories that were specifically positive or
negative, but just to relate memories as they came, they were more likely to retrieve
memories whose valence correlated with the direction of movement. The directedness of

3 Google scientists have just pulled off something similar with Agent (Mnih et al. 2015).

41
the movements involved suggests an important association between memory, movement
and spatial orientation likely to be reflected in shared neural circuitry.

The reuse of spatio-visual circuits for numerical cognition is illustrated by the

spatial-numerical association of response codes (“SNARC”) effect. Here are just two
examples of the SNARC effect (Dehaene et al. 1993): (i) when asked whether a number is
even or odd, subjects respond more quickly with large numbers displayed to their right,
or small numbers to their left; (ii) when presented with a line of neutral symbols (e.g.
XXXXX) subjects fare better at correctly indicating the midpoint than when presented
with small numbers (e.g. 22222), in which case there is a bias to the left, or large numbers
(e.g. 99999), where the bias is to the right. It appears that in these cases a mental number
line running from left to right is being navigated with the help of spatial orientation
circuits (Hubbard et al. 2005).

A growing body of literature within social psychology also supports the case for
reuse. These findings corroborate the influence of physical sensations and relations such
as heat, distance and texture on higher-order processes involved in social cognition
(Williams & Bargh 2008a). Holding a cup of hot as opposed to iced coffee influences the
impressions formed of a target person even when identical information about the target is
given. Those with warm cups gave generally more positive assessments of the target than
those holding iced coffee. In a follow-up study, “those in the warm-coffee condition were
more likely to give their compensation for being in the experiment to a friend (in the form
of a gift certificate), whereas those in the cold-coffee condition were more likely to keep it
for themselves” (Bargh et al. 2010, pp. 267-268). Here we see experiences of physical
warmth directly influencing perceptions of interpersonal warmth. Likewise physical
distance appears to influence feelings of emotional distance and isolation (Williams &
Bargh 2008b). Ackerman et al. (2010) reported cases demonstrating that texture and
weight can affect psychological variables. When a human relations officer holds a heavier
clipboard during a job evaluation, she is disposed to think of the job candidate as more
serious, and overall takes the HR role more seriously. Similarly, negotiation tasks in
which the participants are seated on hardwood as opposed to cushioned chairs induce less
amenability and bargaining flexibility. All these physical-to-psychological effects are at

42
the very least consistent with extensive shared neural circuitry, just as reuse would
predict.

3.3.3 Final thoughts

Despite the extensive and compelling nature of the evidence supporting reuse, not to
mention again the powerful evolutionary considerations in its favour, the case has not yet
managed to convince everyone. Neural activation and imaging evidence on its own is
ambiguous, the skeptics point out, being consistent with multiple neighbouring sets of
neurons that only appear to be reused as a result of the coarse spatial resolution of
contemporary imaging technologies (Anderson 2010, pp. 298-299; 2014, p. 29).
Furthermore

…because neural activation may spread around the brain network, this can lead to
false positives: regions that are activated only as a side effect of their connectivity
and not because they are making a functional contribution to the task under
investigation. (Anderson 2014, p. 29)

This “spreading activation” is what Colin Klein (2010, p. 280) has dubbed a “potential
confound,” and such worries cannot be lightly dismissed. On the contrary, misgivings
about the use of neuroimaging evidence are precisely why converging biobehavioural
evidence (of the kind just cited) will be critical in a debate like this. The more
biobehavioural evidence of functional and semantic inheritance between task domains, the
greater our confidence that the very same neural structures are involved (Anderson 2014,
p. 30). The limitations of neuroimaging technology can thus be overcome by adopting
supplementary research paradigms. An interference paradigm, for example, asks
participants to process two stimuli at the same time. If the processing of these stimuli
draws on shared neural resources, one would expect this to be reflected in performance:
perhaps a slower reaction time as compared to performance on similar tasks that do not
make processing demands on the same neural elements. Thus on the assumption that the
fusiform face area would respond to objects of expertise as well as faces, Gauthier et al.
(2003) predicted that face processing in car experts would be impeded by the presentation
of cars at the same time—and this is just what they found. Here we have evidence of
reuse coming from a research paradigm outside neuroimaging, and none the worse for
that. Later on I cite yet a further type of evidence, this time from single-neuron studies,
43
demonstrating that while concerns over poor spatial resolution and spreading activation
may be legitimate, they can hardly be decisive (see the discussion of “strong context
effects” in § 5.1). The simple fact of the matter is that evidence in support of reuse comes
from many quarters, including from disciplines—such as neurology and
neuropsychology—in which distributed parallel processing models, important precursors
to reuse inasmuch as reuse presupposes them, had been proposed well before the advent
of neuroimaging technology:

One of the earliest and longest running debates in neuropsychology concerned the
question of whether functions of the cerebral cortex are localized in circumscribed
areas or are equally represented throughout the entire cortex. Resolution of this
controversy…slowly came about with the realization that cortical areas do
perform unique elementary functions, but that complex functions require the
integrated action of many areas distributed throughout both cerebral hemispheres.
According to this view, a complex function is a system of interrelated
processes…implemented neurally by a complementary system, or network, of
functionally related cortical areas. (Bressler 1995, p. 289, my emphasis)

Still, even assuming that the general thrust of this hypothesis is correct (as I for one do),
it is not immediately obvious that anatomical modularity is dead, for perhaps it is only in
respect of its functional scope that it stands in need of revision (much as Bressler implies
in the quotation). Moreover, as I suggested earlier in this chapter, outstanding questions
concerning the existence of a dedicated language module remain as acute as ever, and
these are tied in part to an extensive dissociation literature as well as to the concerns over
spatial resolution and neuroimaging I just raised. I turn now to consider these issues, and
begin by inquiring into just what the implications of reuse and neuroplasticity might be
for the modularity of mind.

3.4 SUMMARY

Neural reuse refers to the exaptation of established and relatively fixed neural circuits
without loss of original function/use. Reuse arises over the course of normal development
and evolution. The evidence of this phenomenon speaks most loudly against the idea of
strict domain specificity. It seems that no area of the brain is exempt from redeployment,
with areas of the brain traditionally considered to be among the most domain-specific
(such as sensory areas) also contributing their computational/structural resources to other

44
domains, including those involving language. The evidence supporting reuse takes many
forms, among them evolutionary and developmental considerations, computational
considerations, and the neuroimaging and biobehavioural literature.

45
 4
_____

Modules reconsidered: Varieties of modularity

4.1 A PIVOT TO THE NEUROSCIENCES

Evidence of neural reuse points to an overall picture of the brain that has disruptive
implications for the modularity of mind, particularly for classical varieties of the theory
such as Fodor’s, massive modularity and ACT-R which all posit modules for high-level
cognitive functions or proprietary domains. No doubt many will resist this assessment. I
certainly have sympathy with the tradition of functional decomposition, and shall not in
any case be recommending that we dispense with modules here. Nonetheless such
evidence of reuse as we have clearly does point to “the need for a supplement to business
as usual” (Anderson 2010, p. 249).

The central problem for modularity, at least as it has traditionally been

understood, is that modules talk lends itself most naturally to the analogy of bricks and
mortar, or the assembly of component parts. As an intricately dense network of synaptic
connections, electrical signals and neuromodulatory dynamics, however, the brain is
nowhere obviously organized in this bricks and mortar sort of way, even where it
sometimes proves fruitful to account for neurobiological function in mechanistic
compositional terms (Craver 2007; Bechtel 2008b). The question is whether the bricks
and mortar analogy is so far superseded by the network analogy that there is no longer
any residual value in speaking of modules at all. If the brain is not obviously or even
predominantly an assembly of functional components, surely it would not be

46
unreasonable to hope that any theory having as its target the mind’s functional
organization would adequately accommodate itself to this fact.

The brain’s network structure notwithstanding, metaphors, it seems, die hard,

especially ones freighted with as much philosophical baggage as modularity. It may be
that metaphors are all we have, but if so we are going to need the right ones. To those
reluctant to give up on the modular perspective I hope my own recommendation of a
substantial yet cautious reform may offer some consolation. My proposal is simple—that
we recalibrate our notion of modules in deference to what currently passes for a module
in contemporary mainstream neuroscience. Cognitive scientists and philosophers whose
work is attentive to the neurosciences already think in these terms, and it is not hard to
appreciate why: when it comes to modularity, which concerns the functional organization
of the mind, psychological theorizing is even more constrained by issues of
implementation than might generally be the case. For some reason, however, many
philosophers continue to talk about modules in a manner conveying either ignorance of
what neuroscience has to say about the structure of the brain or else a breezy indifference.
A reorientation towards neuroscience entails a shift of focus away from understanding
modules as unimodal high-level cognitive mechanisms towards a conception of modules
as metamodal (i.e. reusable) nodes subserving exiguous low-level subfunctions. I argue
that a module built on this pattern, sometimes called a “brain module,” can serve as an
appropriate revisionary benchmark for cognitive scientists and philosophers of
psychology still wedded to the idea of classical modules. Notice, however, that this
proposal also entails a certain agnosticism regarding the prospects of modularity in the
long run. Being sensitive to developments in the neurosciences means being willing to
part with long-cherished notions if needs be. It so happens that further evidence of neural
reuse presented in the next chapter may necessitate a more profound shift away from
traditional modules than the one I am currently recommending. Hence I am urging a
recalibration in the face of developments which will either, if all goes well, allow us to
safeguard a respectable (though revised) notion of modularity, or, should things not turn
out so well, undermine its rationale comprehensively—this is where the real battlelines
are being drawn. Later on I shall suggest one way that we might usefully conceptualize
the issues presented by these developments. Still the broader point remains: if modules

47
exist at all—a question on which it pays not to be dogmatic one way or the other—they
will not resemble the modules of classical cognitive science.

In the next section I provide a rough sketch of the varieties of modularity one
might expect to come across in the cognitive sciences. In this section I also defend what I
take to be the sine qua non of modularity, namely, functional dissociability. This will be
important in heading off an obvious objection to the argument I am making here: that
modules can always survive qua abstract, high-level functional “systems.” I follow this
section with a basic account of the brain module. The next chapter pursues at greater
length the central question of this thesis—whither modularity?—in the hope of
demonstrating why neural reuse points us in the direction of something like the brain
module.

4.2 VARIETIES OF MODULARITY

4.2.1 Themes and trends

The nineteenth century phrenologists were probably the first to emphasize the
specialization of brain functions. Gall and Spurzheim (1835) hypothesized “about thirty-
five affective and intellectual faculties” localized in distinct regions of the brain. As almost
everyone knows, however, they got the details horribly wrong, for they fallaciously
assumed that the activity of a cortical faculty would be reflected in its size, and that its
size in turn would be reflected in the relative prominence of cranial bumps. This led them
to endorse the pseudoscientific practice of gauging personality from the shape of a
person’s skull. Wrong though they most assuredly were in this respect, the idea that brain
function can be mapped to local structure was not itself a bad idea. It soon received
empirical support in the work of the neurologists Gustav Fritsch, Eduard Hitzig, Paul
Broca and Carl Wernicke—Broca and Wernicke being of course the first to discover the
so-called “language” areas of the brain (Bridgeman 2010). Indeed by the end of the
nineteenth century, the idea was well on its way to becoming the equipment of every
working scientist in the field. In fact “the notion of cognitive function being subdivided
and routed to different regions of the brain has remained a central idea in neural science

48
and a fundamental principle in the clinical practice of neurology” (Pascual-Leone &
Hamilton 2001, p. 431).

Corresponding to a rough division between mind and brain, one may trace the
course of two distinct but parallel traditions originating in the work of these nineteenth
century neurologists. The first is a structuralist tradition whose methodology, guiding
assumptions and theoretical concerns are predominantly biological (i.e. neurological and
anatomical). From a certain point of view, Fodor’s archetype could be said to fall broadly
within this tradition—notwithstanding the subordinate and strictly dispensable role
played by structural properties in his overall account (Anderson & Finlay 2014, p. 5;
Fodor 1983, pp. 71, 98-99; Coltheart 1999)—as may both the neural network graph-
theoretic module (see § 4.2.2) and neuroscience “brain module” (see § 4.3), which I come
to shortly.

An alternative approach investigates questions of cognitive architecture from the

standpoint of a classic computationalist or functionalist. In the guise of evolutionary
psychology or “massive modularity,” for example, it “retains the Fodorian focus on
computation, and with it a focus on the algorithmic (or heuristic) efficiency of purported
psychological solutions to adaptive problems such as food choice, mate selection, kin
identification and cheater detection” (Anderson & Finlay 2014, p. 5). It does not,
however, entail specific commitments about implementation beyond those required for
functional independence.1 (See Sternberg 2011, pp. 158-159 for an overview.)

These two (ideally) complementary approaches to the mind/brain are reflected

again in the central assumptions underpinning much of the effort within
neuropsychology, cognitive neuropsychology and cognitive neuroscience (Bergeron
2007). Bergeron calls these the “anatomical modularity assumption” and the “functional
modularity assumption.” Recall that in Chapter 1 we provided a general definition of an

1Defenders of massive modularity also part company with Fodor’s “central”/“peripheral” distinction.
Fodor’s hypothesis is that only peripheral systems are likely to be modular “to some interesting extent”
(1983, p. 37), i.e. sensory input and motor systems. Proponents of massive modularity think that the central
systems will be modular too, i.e. those involved in higher perceptual function, belief-fixation and inferential
reasoning (Sperber 1994; 2002; Carruthers 2006; see also Barrett & Kurzban 2006 and Prinz 2006 for
reviews).

49
anatomical module. It is worthwhile restating this definition in such a way as to reveal
more clearly its relation to a “functional” module. The functional modularity assumption

is the idea that the architecture of human cognition largely consists in a

configuration of cognitive modules, where a “module” is roughly defined,
following Jerry Fodor (1983), as a domain specific, innately specified, and
informationally encapsulated system....What this means is that human cognition
can be decomposed into a number of functionally independent processes, and that
each of these processes operates over a distinct domain of cognitive information.
Moreover, since these processes are brain processes, to hypothesize that the
capacity to do A and B depends on two distinct cognitive modules—one
responsible for the capacity to do A and the other responsible for the capacity to
do B—is to hypothesize that the brain processes cognitive information related to
A separately from the way it processes cognitive information related to B….

What makes the A module/process distinct from the B module/process is their

functional independence, the fact that one can be affected, in part or in totality,
without the other being affected, and vice versa. (Bergeron 2007, pp. 175-176)

The anatomical modularity assumption, then,

is the idea that the cognitive modules which compose cognition (or at least most of
them) each reside in some specific and relatively small portion of the brain….The
anatomical modularity assumption is in fact the functional modularity assumption
plus a claim about the implementation of functionally modular processes in the
brain. (Bergeron 2007, p. 176, my emphasis)

Stripped to their essentials, functional modularity implies functional dissociability, while

anatomical modularity implies both functional dissociability and neural localization. As I
argue below, functional dissociability—functional modularity pure and simple—
represents the essence of any modular account worthy of the name.

What I have so far failed to mention, though it will in fact be crucial to

appreciating the implications of neural reuse, is that cognitive modules have been
generally postulated to account for higher level or gross cognitive functions, i.e. for the
sorts of psychological capacities that might appear in the ontologies of cognitive
psychology. Even if one restricts one’s gaze to the history of the structuralist/neurological
tradition, one will not be surprised to learn that, in the main, the project of mapping
function to structure has proceeded with a fairly coarse taxonomy of psychological
capacities in hand. The phrenologists, for their part, merely translated the categories of

50
Thomas Reid’s faculty psychology onto a plan of the skull (acquisitiveness, friendship,
sagacity, cautiousness, veneration, etc.) (Poldrack 2010, p. 753). Broca’s postulation of a
“language” area associated with motor aphasia, though no doubt empirically better
supported than Gall and Spurzheim’s assumptions, hardly served to sharpen the focus on
what the brain itself is actually doing when facilitating speech; for “what warrants the
thought that [such] characteristics [as those found in faculty psychology] will be useful to
structuring the neuroscience of behaviour and divide the brain at its functional joints?”
(Anderson 2014, p. xvi). Consider Russell Poldrack’s illuminating reductio ad absurdum
(cited in Anderson 2014):

Imagine that fMRI had been invented in the late 1860s rather than the 1990s.
Instead of being based on modern cognitive psychology, neuroimaging would
instead be based on the faculty psychology of Thomas Reid and Dugald Steward,
which provided the mental “faculties” that Gall and the phrenologists attempted
to map onto the brain. Researchers would…almost certainly have found brain
regions that were reliably engaged when a particular faculty was engaged, …[and]
Gall and his contemporaries would have taken those neuroimaging results as
evidence for the biological reality of his proposed faculties. (Poldrack 2010, p.
753)

What reasons have we for imagining that the taxonomies of modern day psychology will
fare any better in carving the brain at its true functional joints? Clinical evidence of
dissociations aside (about which I shall have more to say later), widespread evidence of
neural reuse strongly suggests that attempts which seek to impose upon the brain a set of
categories devised (largely) autonomously of the brain, and moulded from a wholly
different set of considerations from those guiding brain science generally, are doomed to
repeat the same basic phrenological mistake. What is needed is “the development of
ontologies that let the phenomena speak on something closer to their own terms”
(Anderson 2014, p. xvii).

The structuralist tradition in fact does admit of some exceptions to this

questionable trend in what might even be seen by some as a clear premonition of neural
reuse theories. As Bergeron’s (2007) helpful discussion reminds us, Carl Wernicke’s
postulation of a sensory speech area, often wrongly dubbed the “language comprehension
area,” obscures the fact that Wernicke himself was “very resistant to postulating any
cerebral centers beyond what he referred to as the ‘primary’ (motor and perceptual)

51
‘psychic functions’ ” (2007, p. 184). Wernicke could well be credited with the elaboration
of an entirely original approach to the structure-function relation in the brain. On this
approach, only the sensory and motor functions are allocated distinct and dedicated
neural anatomy. Higher psychological functions such as those implicated in language
production and comprehension are supposed to depend on the interactions of these low-
level sensory-motor systems. This arguably anticipates modern theories of reuse which
predict that higher cognitive functions resolve in the interactions of lower level elements.
Bergeron certainly thinks so, and even suggests that Wernicke must have been operating
with an implicit understanding of the difference between a cognitive working and a
cognitive use, the distinction which, as we saw in Chapter 3, Anderson made central to
his original presentation of the massive redeployment hypothesis. If Bergeron’s
conjecture is correct, Wernicke’s great methodological innovation—what set him apart
from the phrenologists and even his predecessor Paul Broca, for example—consisted in
his cautious reluctance to infer cognitive working (i.e. essential functional contribution
across all task categories, considered in isolation of neural context) from cognitive use (i.e.
high-level cognitive function), an inference obviously susceptible to Poldrack’s reductio.

In the same vein, the father of modern neuroscience and champion of the neuron
doctrine, Santiago Ramón y Cajal, “was decidedly not a supporter of either the definition
of psychological ‘faculties’ or their assignment to discrete, localized neural ‘organs’ ”
(Anderson 2014, p. xv):

In [his] view, brain function is to be understood in terms of a hierarchy of

reflexes, in the most sophisticated instances of which one responds not just to
external but also to internal, and not just to current but also to stored stimuli….In
such a brain there can be no region for circumspection or poetic talent, for
although a particular sensory experience or association may be stored in a
particular place…the behavioral characteristics of the organism are realized only
by the fluid activity of the whole system in its environment. (Anderson 2014, p.
xv-xvi)

The idea that specific circuits could be cued by various stimuli across both internal and
external environments is a tolerably clear presage of the metamodal hypothesis of brain
organization which we encountered briefly in Chapter 2 and underwrites the possibility of
neural reuse. (I revisit the metamodal hypothesis in more detail in the next chapter, as it
bears greatly on the questions facing us there.)

52
 4.2.2 Graph theory and network neuroscience

There is another usage of the term “module” that one often comes across in the literature.
It is perhaps testament to the immense versatility of modularity that it has descriptive
utility well beyond the confines of cognitive science. Modules play an important role in
fields as diverse as developmental and systems biology, ecology, mathematics, computer
science, robotics and industrial design. One interesting application of the term occurs in
the study of networks, and neural networks in particular. Unfortunately, there is a danger
of confusion here, because the network concept is significantly looser than the classical
one in cognitive science. Thus it sometimes happens that different researchers, all of
whom work in the cognitive sciences broadly speaking (including brain science), refer to
“modularity” but mean different things by it.

A network is any organization with a weblike structure. The Internet, airline

routes, food webs and electrical grids spring immediately to mind, but these are only the
most obvious examples among a great variety of phenomena displaying network design,
including genetic regulation and protein interaction (Bullmore & Sporns 2012; Caldarelli
& Catanzaro 2012, pp. 23-25). Networks manifest a number of important universal
properties (Caldarelli & Catanzaro 2012, pp. 4-5). At the most elementary level, all
networks comprise a collection of nodes (or “vertices”) and the various connections (or
“edges”) between them (see Fig. 1). In a map of airline routes, for example, a single
airport would be represented by a node and the route between any two of them by an
edge. Because the focus of attention is the global structure of interactions between nodes,
rather than the individual nodes themselves, the basic representational vehicle can be the
same in every case, namely a graph depicting nothing more than these nodes and their all-
important interconnections (Caldarelli & Catanzaro 2012, pp. 4, 12; Anderson 2014, p.
12). In graph theory, a “module” is defined as a community of “densely interconnected
nodes” where “the existence of several [such] communities is characteristic of [a] modular
[network]” (Bullmore & Sporns 2012, p. 342; Caldarelli & Catanzaro 2012, pp. 89-90)
(Fig. 1). In network neuroscience specifically, network models take the form of neural
coactivation graphs, where modules are identified as communities of nodes that are
functionally coactive (see below). In the context of neural networks, then, “modularity

53
refers to the existence of multiple communities of neurons or brain regions as defined by
patterns of [functional] connectivity” (Bullmore & Sporns 2012, p. 342).

Figure 1. Nodes, edges, modules

and hubs in a network. Nodes are
sometimes also called vertices.
Source: Bullmore & Sporns 2012,
p. 342.

The point is explained very simply by Caldarelli & Catanzaro in connection with
the importance of functional magnetic resonance imaging (fMRI):

When humans perform an action, even one as simple as blinking, a storm of

electrical signals from the neurons breaks out in several areas of the brain. These
regions can be identified through techniques such as functional magnetic
resonance. Through this technique, scientists have discovered that different areas
emit correlated signals. That is, they show a special synchronization that suggests
that they may influence each other. (Caldarelli & Catanzaro 2012, p. 27)

Furthermore

These areas can be taken as nodes and an edge is drawn between two of them if
there is a sufficient level of correlation. Also at this level, the brain appears as a set
of connected elements [i.e. “modules”]. Each action of a person lights up a
network of connected areas in the brain. (Caldarelli & Catanzaro 2012, p. 27)

That is, the neuroimaging data resulting from a functional connectivity analysis can be
represented as a graph—a neural coactivation graph—in which nodes represent
individual brain regions and edges denote the likelihood of coactivation between two
nodes during a particular task (Anderson 2014, p. 12) (see Fig. 2).

Why should this prove instructive for cognitive architecture? It turns out that the
abstract topological features of these neural coactivation graphs frequently (if only
roughly) track the functional taxonomies of cognitive psychology, cognitive
neuropsychology and the computationalist/functionalist tradition more generally

54
(Anderson 2010, p. 303; 2014, p. 42). This sense of the word “module” therefore seems as
if it might have a natural affinity with the modules to which philosophers of psychology
have become accustomed. But closer inspection shows this to be a tentative link at best.

Figure 2. Each community of nodes forms a module that tracks a higher level cognitive function such as
attention or semantics. Source: Anderson 2016, p. 4.

Firstly, being in effect sets of reusable (i.e. domain-general/task-selective) nodes,

these graph-theoretic modules are not your typical dissociable ones (although see Pessoa
2016 for discussion); nor, for that matter, are they intended to track encapsulation,

55
domain specificity, automaticity or the half dozen other features typically ascribed to
modules within the computationalist tradition (Stanley & De Brigard 2016). Quite
simply the usage here is sui generis. Secondly, while there no doubt is a standard and
more orthodox usage of the term “module” in neuroscience (one which moreover does
offer some support to the classical conception from cognitive science, as I discuss later),
its meaning is in fact much closer to what is represented by the nodes of a coactivation
graph than by the communities of nodes in such a graph (see e.g. Pascual-Leone &
Hamilton 2001, p. 443; Pascual-Leone et al. 2005, p. 396; Caldarelli & Catanzaro 2012, p.
27; Fedorenko & Thompson-Schill 2014, p. 120, 121; Zador 2015, p. 44). That is, the
standard sense of “module” in neuroscience trails far more closely the idea of small
individual brain regions with discrete subfunctional profiles than it does the idea of high-
level functional composites. The anomaly results from the fact that network techniques
were developed independently of neuroscience, and with a distinctive usage and
vocabulary. When network methods were eventually adopted by neuroscientists, an
idiosyncratic usage was introduced into a discipline that already had a fairly settled
meaning for the term “module.” In neuroscience, “module” typically refers to a cortical
column (akin to a node in the coactivation graphs above), and this, as we shall see further
in §§ 4.3 and 5.1, is a twentieth century refinement of the anatomical module within the
structuralist tradition.

4.2.3 Separate modifiability as the touchstone of modularity2

A common objection to accounts of cognitive architecture which downplay or question

the modular hypothesis is that modularity has not been given due credit for the uniquely
versatile concept that it is, and that the dissenters have simply fettered themselves with an
impossibly narrow and needlessly structuralist conception of cognitive architecture that is
unwarranted in all the circumstances (the circumstances being the Cognitive Revolution,
the fact that no one seriously denies that the mind has a rich internal structure, the
unquestionable boon of functional decomposition as an effective research strategy in the
cognitive sciences, etc.). Jungé and Dennett (2010, p. 278) appear sympathetic to this
point of view: “A software theory of massive modularity—programs evolved to serve
particular adaptive functions within brains—without commitments about implementation

2 Material from this section appears in the journal Philosophical Psychology (see Zerilli 2017a).

56
(unlike anatomical modularity) could survive [the evidence of neural reuse] largely
untouched.” At issue here is whether a nondissociable system of some variety could be
regarded as modular—whether, say, a language or norm acquisition device comprised of
very many smaller domain-general neural regions could in some sense be a module.3
Against this suggestion is the claim that functional dissociability ought to fix a definite
threshold beneath which a system cannot be regarded as modular. Here I shall contend
for the latter view.

Recent developments in neuroscience have no doubt added to the lustre of the

“system” module, as I shall call it, and even encouraged the view that such modules
represent what was always the most important contribution of modular theories to our
understanding of the mind (see below). But actually the system module has been around
for a long time. Its fortunes can nowhere be more illuminatingly charted than in the
annals of generative grammar. Generative grammarians are notorious for prevaricating on
the issue of linguistic modularity—one can easily locate passages which would suggest the
modularity in question is anatomical or at the very least functional (Chomsky 1975, pp.
40-41; 1980a, pp. 39, 44; 1988, p. 159; 2002, pp. 84-86; Pinker & Jackendoff 2005, p. 207;
Fitch et al. 2005, p. 182; Collins 2008, p. 155) and others where what they seem to have in
mind is little more than a “domain of inquiry”—“[t]he view that internal cognitive
systems can fruitfully (for purposes of successful theory-construction) be studied
independently of other such systems” (McGilvray 2014, p. 235; Chomsky 2005, p. 5).
Notice that the module-as-domain-of-inquiry very effectively neutralizes the sting of
neuroscientific evidence, since there is really no evidence that neuroscientists can adduce
against the existence of such a module (a point to which I return below). Indeed the
system module is frequently endorsed by playing down the significance of
implementation and emphasizing its “methodological value as a research heuristic”
(Badcock et al. 2016, p. 11; see also Scholl 1997). But let us return to the other theme of
this section, the notion of dissociability.

In a straightforward sense, a system is dissociable if it is functionally specialized—

if it can (in principle) be modified without directly impeding the operation of a

3In § 5.2 I consider whether it is possible for systems consisting of shared domain-general parts to be
functionally dissociable. This is the same as asking whether high-level cognitive functions could persist as
functional modules. For now we can assume the answer is no.

57
comparable system.4 More precisely, if a neural system n consisting of neural primitives
{p1, p2, p3 … pn} contributes some specific and functionally discrete operation f such that
all (or most) elements of the set {p1, p2, p3 … pn} are dedicated to f, n will be dissociable.
On this understanding, a speech production centre will be dissociable if its impairment
has no direct effect on any system “considered with the same grain of analysis”
(Carruthers 2008, p. 295) (e.g. numeracy, rhythm, speech comprehension, episodic
memory, IQ, etc.) even though it might ramify to compromise a higher level functional
system that draws upon the speech production centre for processing (e.g. singing,
signing, etc.). In the context of neural reuse, we may presume that a working’s
impairment will ramify to all higher level functional composites in which it plays an active
role; and yet so long as no other working is directly put out by such an intervention, the
working remains sufficiently discrete to be regarded as dissociable. (Whether brain
regions as small as workings are truly dissociable in this sense is another question. I take it
up in Chapter 5.)

Notice that when spelled out in this way—and all I have done is follow through
with the logic of dissociability as it is commonly understood (see e.g. Carruthers 2008, p.
258; Coltheart 2011, pp. 227-228)—the requirement could be thought to lose much of its
explanatory power. For what it entails is that the smaller and more functionally
promiscuous a neural system gets—remembering that neural reuse itself implies that the
typical brain region will be both extremely small and highly versatile—the more difficult
to quarantine the effects of regional impairment, since those effects will presumably
ramify to all affected distributed systems. An evolutionary psychologist might allege that
nothing theoretically significant can follow from the fact that a tiny brain region is
dissociable if its impairment will disturb the operation of many higher level cognitive
systems. It is only when modules directly implement high-level cognitive functions (e.g.
sentence parsing, cheater detection, face recognition and the like) that the effects of
modification can be contained in a way that makes dissociability an important constraint
on cognitive theory. For then evolution itself can have a clear role to play in shaping
cognitive systems by selectively modifying brain regions in a way that does not
reverberate detrimentally across distributed systems. This indeed was thought to be a

4Its modification may of course indirectly impede a comparable “downstream” system, i.e. one at the
receiving end of its efferent projections.

58
major argument in favour of modularity—the neat solution it offered to the so-called
evolutionary “debugging” problem (Cosmides & Tooby 1994). By contrast any account of
modularity in which modules come out as small and promiscuous is an account of
modularity which no longer promises to solve the debugging problem. And (it may be
alleged) any criterion of modularity which casts modules in such a diminutive role cannot
be considered especially salient.

Now I am defending dissociability as a criterion of modularity. My position must

therefore seem a little curious, for am I not by defending dissociability actually defending
the wrong sorts of modules—given the sorts of modules that this criterion delivers if the
redeployment hypothesis is correct? I can certainly see how an evolutionary psychologist
would be puzzled by my position. But, as I shall explain later, I do not think the
evolutionary psychologist’s reasoning here is persuasive—frankly, the sorts of modules
she is after are very unlikely to be found anywhere beyond the most primitive domains,
and the search for them at all reflects a misunderstanding of the brain and its evolution:
the debugging problem is not a deep one. Like it or not, therefore, it looks as if we are
going to have to rest content with a diminutive role for modules—which may not be such
a bad thing anyway. For while dissociability may not ultimately meet the desiderata for a
theory of evolutionary psychology, it ought to safeguard a respectable threshold for
modularity nonetheless. It furnishes a kind of cognitive movable type for the mind, and
mechanisms that can support robust laws, generalizations and predictions (e.g. “forward”
inferences from cognitive tasks to brain areas) (Burnston 2016). If

For a given neural area A, there is some univocal description D, such that D
explains the functional role of A’s activity whenever A functions

it should be possible to formulate a theory tokening A providing “functional descriptions

that apply over a range of instances of functioning,” and “functional explanations in
particular contexts that are relevant to contexts not yet explored” (Burnston 2016, pp.
529, 531). This would be a “very powerful theor[y] in terms of generalizability and
projectability” (Burnston 2016, p. 531).5

5Brain regions that are domain-general in the way envisaged by theories of neural reuse may of course
ultimately prove not to sustain completely generalizable and projectable accounts of local function. The
ability of a brain region to maintain a set of stable input-output functions, and hence to be truly dissociable,
may be compromised by the effects of the neural network context. I pursue this topic in Chapter 5.

59
 So what, then, of Jungé and Dennett’s suggestion? The problem, as I see it, is that
it confuses modularity with faculty psychology more generally, and so reduces it to a
platitude. Being neither controversial, falsifiable, or particularly interesting, the system
view fails to live up to the theory’s venerable reputation. On such an expansive definition,
who would not emerge as a defender of modularity? Certainly few theorists in the
cognitive neurosciences would deny the utility of functional decomposition as an effective
research strategy (Prinz 2006; Piccinini & Craver 2011; Boone & Piccinini 2016).6 And of
course a high-level cognitive system composed of shared neural elements might well
exhibit natural kind properties, such as a systematic set of procedures for dealing with
typical inputs (Chomsky 1980a; 2006; Pinker 1997). But it is difficult to see how such a
definition could have any substantively worthwhile theoretical upshots, certainly of a kind
that could possibly justify the enormous effort spent in advancing modularity as some
sort of solution to a deep and longstanding set of issues. On this weak view, what would
the modularity of cognition explain about cognition beyond the simple fact that the mind,
too, may be investigated using the techniques of natural science (i.e. “divide and
conquer” works here too)? If the answer is “not much,” this cannot be a good account of
modularity—assuming that by “modularity” we mean a substantive doctrine. On the
weak construal, modules turn out to be little more than fruitful perspectives on the mind,
the mind considered from this or that particular point of view, say, the point of view of its
linguistic capabilities, its pitch discrimination capabilities, its problem-solving
capabilities, and so on (in principle without limit). Such perspectives unquestionably give
us useful entry points into what would otherwise be intractably complex, and allow us to
figure out what it is that the mind actually does. But it is hardly surprising that a targeted
coming-to-grips with a complex object should yield significant insights. The same
strategy is familiar in one form or another in virtually all domains of rational inquiry, be
they physical, chemical, biological, psychological or otherwise. That “science works here
too” I take not to be an interesting claim, if it comes to that, because it does not so much

6 Decomposability and modularity do come apart. Boone and Piccinini (2016, p. 1524) outline “a
mechanistic version of homuncular functionalism, whereby higher-level cognitive capacities are iteratively
explained by lower-level capacities until we reach a level in which the lower-level capacities are no longer
cognitive in the relevant sense.” While this might entail modularity for some lower-level elements (they do
not say as much), it does not entail modularity for higher-level elements composed predominantly of shared
parts (indeed the word “modularity” or “module” appears nowhere in their paper): see McGeer’s (2007)
helpful discussion of the cognitive neuropsychologist’s understanding of modularity. Prinz (2006) is
actually explicit that so long as the units of decomposition do not exhibit the properties associated with
Fodorian modularity, we should proceed with decomposition but abandon the label of modularity. See my
remarks, below, for further clarification of this point.

60
as provide a theory of cognition at all: if anything it says more about science than it does
about cognition. Furthermore it is not entirely clear that the behaviourists would have
spurned the sort of modules in view here. What they denied was the existence of sui
generis principles, or the computational/architectural/modality independence of certain
capacities (e.g. language), which I regard as evidence of true modules. They would not as
a rule have denied that a partition of their subject matter could lead to interesting results.
Recall the title of Skinner’s Verbal Behavior—I think it is fair to say that Skinner sought
quite literally to explain the language faculty, albeit in terms of general associationist
learning mechanisms, and is this not nearly comparable to the system sense of a module
now under discussion? The behaviourists may have offered a shallow theory of human
capacities, but even it did not appear to preclude modules in this sense (see e.g. Chomsky
1979, pp. 49-51). Nor for that matter is there any logical reason why a connectionist or
holist would have to rule them out either. Contemporary PDP models of cognitive
architecture in fact do have a sort of generic componential motivation behind them
(O’Reilly 1998; Jilk et al. 2008).

It is worth being clear about exactly why system modularity fails the test of
“interesting.” It is easy to be misled here by the genuinely “interesting” results which
have been achieved as a consequence of adopting the system view, i.e. by what has been
learned about distinct domains of psychology as a result of iterative functional analysis
(task analysis, decomposition, “boxology,” etc.). The system module’s notable successes
as well as its historical association with the computational theory of mind and the view of
the mind as richly and intricately structured are apt to lead to an exaggerated estimate of
its true significance. Any theory of the mind pitched at the level of faculties (or analyzable
parts, components, units, etc.)—as modularity most assuredly is—must tell us what it
does about the mind through what it tells us about the faculties (or whatever the relevant
units of analysis happen to be). If it does not speak “through the faculties,” as it were, it
cannot so much as count as a faculty psychology, since the properties of the mind to
which a faculty psychology brings our attention are, in the first instance, properties of the
mind’s faculties. This point is at once obvious and yet so readily overlooked that it needs
to be emphasized. That the mind is richly structured, that the mind is a computer, that
the mind obeys laws exhibiting a clear mathematical structure, and so forth—these
statements, if they are true, are true of the mind generally, meaning that it ought to be

61
unsurprising if the divisions of the mind are correspondingly rich, intricate,
computational, systematic, and so on. None of these properties attach to faculties per se.
Moreover, learning that the language faculty has such and such features, or that vision
operates in this or that fashion, need not tell us much about faculties qua faculties either,
as against telling us about this or that particular faculty. Thus neither general claims
about the mind associated with the Cognitive Revolution, nor specific claims about
specific faculties, hard won though these insights may have been, automatically get
reckoned as among the distinctive insights marking out a truly general theory of faculties,
which is after all what a faculty psychology aims to be. Contrast such claims with those of
a well developed faculty psychology (e.g. Fodor 1983). The roster of properties associated
with Fodorian modularity (domain specificity; encapsulation; shallow, fast and mandatory
processing; hardwiredness; etc.) do not amount to a list of properties pertaining to the
mind generally, nor to specific faculties considered independently, but to all faculties qua
faculties. This is what made his theory interesting. So as easy as it is to roll the system
module in the glitter of the Cognitive Revolution, a frank assessment of this module
demands that we isolate clearly what it is the theory which posits such modules says about
the mind at the level of faculties—and when we do this, I maintain, we will be hard put to
find anything that would not heartily be conceded by anyone who believes in the power of
science (be they classical modularists, connectionists, holists, and, as I suggested,
probably even behaviourists mutatis mutandis). Furthermore, lest it be thought that the
very idea of functional decomposition can underwrite the system view—for one must
admit that decomposition proceeds in a curious fashion in the context of computational
systems, namely the execution of subroutines by homunculi, surely a nontrivial design
feature of such systems—it need only be pointed out that homuncularity is not the same
thing as modularity. Careful psychologists have always understood the difference, and
that modularity is really a special type of homuncularity (Mahon & Cantlon 2011, pp.
149-151), just as homuncularity is a special type of decomposition (van Gelder 1995, p.
351). It is interesting to observe in this connection that David Marr, one of the chief
architects of the computational theory of mind, did not see computationalism (and
therefore, we may surmise, homuncular functionalism) as providing a free pass to his
“principle of modular design.” Modularity seems for Marr to be an added feature that
some computational systems, for largely heuristic reasons, might be thought to possess:

62
 Any large computation should be split up and implemented as a collection of
small sub-parts that are as nearly independent of one another as the overall task
allows. If a process is not designed in this way, a small change in one place will
have consequences in many other places. This means that the process as a whole
becomes extremely difficult to debug or to improve, whether by a human designer
or in the course of natural evolution, because a small change to improve one part
has to be accompanied by many simultaneous compensating changes elsewhere.
(Marr 1976, p. 485)

So although homuncularity is not as generic as “mere decomposition,” it is nowhere near

as important a principle as modularity either. Accordingly (and for additional reasons I
canvass below), we should withhold the more serious designation from generic system
subcomponents and procedures that are nondissociable.

So far I have said nothing about two important features of classically modular
systems, domain specificity and informational encapsulation. Can they get the system
module over the line? Actually the question itself is incoherent. Consider that once a
module is allowed to consist of shared parts, it can no longer be domain-specific, except
perhaps in an abstract sense. This is because the “module” will be sensitive to potentially
many domains, since its parts are presumably domain-general (see below, § 4.3). Put
another way, domain specificity7 requires a functionally integrated unit that can respond
to specified inputs. While the component modules of a composite consisting of shared
parts would be functionally integrated, it is not obvious that the composite itself would
be, although it might be said to have a sort of ad hoc integrity when in use. Notice also
that a composite is unlikely to be informationally encapsulated “precisely because in
sharing parts [it] will have access to the information stored and manipulated by [other
high-level systems]” (Anderson 2010, p. 300). Anatomically distributed and overlapping
brain networks simply must share information on some level (Pessoa 2016, p. 23).
Lacking both of these properties, then, one or the other of which has been considered
definitive (Coltheart 1999; Fodor 1983, p. 71), its postulation does not quite serve the
purposes many would assume. One might have supposed that the system module could
be more strongly motivated if at least it had the property of either domain specificity or
encapsulation (in a concrete and unambiguous sense). And yet just because it is a

7 Of whatever variety—strict or formal (see §§ 2.4.3 and 5.1).

63
composite it can be neither truly domain-specific nor (in all likelihood) informationally
encapsulated.

In something of a reductio, then, the system view of modularity leads only to the
claim that the mind can do different things at different times. Certainly a more ambitious
and theoretically interesting claim than this is available, namely that the mind can do
different things at the same time; but as far as we know this requires functional
specialization, i.e. separate moving parts (real modules), since the prospect of neural time-
sharing appears low (see § 7.5 on the time-sharing problem). The pervasiveness of
cognitive interference effects and processing bottlenecks in stimulus-rich environments
that impose overwhelming attentional demands are enough to make this clear (Anderson
2010, p. 250). In short, either these debates are trifling, or the claims at stake are more
adventurous than the system view permits. Here I shall presume that the more
adventurous reading is correct, and that, in any event, functional dissociability really
ought to be considered the sine qua non of modularity.

Bear in mind also that in the context of cognitive neuropsychology, modules have
been defined largely by reference to what the dissociation evidence has revealed, i.e. “on
the basis of the specific behavioral effects of brain lesions” (Bergeron 2007, p. 177).
Bergeron calls this inferential strategy the “functional modularity inference.” Basically,
“the presence of highly selective cognitive impairments (dissociations) like prosopagnosia
and various linguistic processing deficits suggest [sic] the functional independence of at
least some cognitive processes,” and this in turn licenses the postulation of functionally
independent modules subserving those processes (Bergeron 2007, pp. 176, 177;
Gazzaniga 1989, p. 147). The fact that brain lesions are often also relatively localized
suggests that such modules reside in a “specific and relatively small portion of the brain”
(the “anatomical modularity inference”) (Bergeron 2007, p. 176; Gazzaniga 1989, pp. 947,
950). Make no mistake, the legitimacy of these inferences is hotly contested, since noisy
dissociations are compatible with a system’s being dissociable, and clean dissociations
compatible with a system’s being substantially nondissociable. In the first instance, “there
are a variety of reasons, well explored in the neuropsychology literature, for which lesions
to brain systems can produce noisy rather than clean patterns of breakdown even when
the systems required to complete the task are modular” (Barrett & Kurzban 2006, p. 642).

64
A good example would be a focal lesion at the border of two adjacent modules—the
breakdown would not be clean, and yet the two systems would be modular. In the second
instance, even perfect (or “double”) dissociations cannot conclusively establish that the
affected systems are modular, for a lesion might only compromise a small isolable
component of an otherwise highly interpenetrative circuit. Damage to this component
might result in the system depending on that component being independently impaired,
but it does not follow from this that the system would be functionally dedicated (although
admittedly it would be dissociable at the margins). In light of this, it may come as
something of a surprise to be told that these arguments

have failed to deter theorists from employing either of the inferential strategies.
Indeed, the functional modularity inference continues to be one of the most
common approaches among cognitive neuropsychologists for inquiring about the
structure of cognition. Similarly, the recent cognitive neuroscience literature
abounds more than ever with cases involving the use of the anatomical modularity
inference. (Bergeron 2007, p. 177)

But what the persistence of these inferences bears witness to is the fundamental role that
dissociation evidence plays in the search for modules, and that functional specificity itself
continues to be the lodestar for deciding upon whether and if so to what extent the mind
is modular within the major disciplinary fields concerned with this question. As we have
seen, the same assumption underwrites evolutionary psychology and massive modularity,
the central claim of which is that the mind is predominantly composed of parts selectively
shaped by evolutionary pressures (Carruthers 2006). As two prominent evolutionary
psychologists state their position (Barrett & Kurzban 2006, p. 630): “…we intend an
explicitly evolutionary reading of the concepts of function and specialization: modules
evolved through a process of descent with modification, due to the effects they had on an
organism’s fitness.” This view predicates the existence of systems which, though perhaps
spatially extended and neurally interspersed, are dissociable in principle:

Psychologists generally agree—as do we—that because cognitive architecture is

instantiated in brain architecture, the two will be isomorphic at some
level….However, at a larger, macroscopic level, there is no reason to assume that
there must be spatial units or chunks of brain tissue that neatly correspond to
information-processing units. An analogy might be to the wiring in a stereo, a
computer, or other electronic system: Individual wires have specific functions, but
at the level of the entire machine, wires with different functions might cross and
overlap. For this reason, removing, say, a three-inch square chunk from the
65
 machine would not necessarily remove just one of the machine’s functions and
leave the rest intact. In brain terms, it could be, and probably is, that macroscopic
regions of brain tissue include neurons from multiple information-processing
systems with multiple functions. (Barrett & Kurzban 2006, p. 641)

And of course other contemporary models of cognitive architecture, such as the

successful ACT-R model, also posit the existence of independently modifiable
subsystems of the brain.

When it comes to clarifying just what makes modularity interesting, one final set
of considerations may be suggestive. While terminological nuances can hardly be decisive
in an area like this, I think it is no coincidence that massive modularity bottoms out in
claims about the separate modifiability of functional components. This is because the very
word “module” evokes images of movable parts that can be assembled and reassembled in
a variety of distinct combinations, and which may be affected independently of one
another. If all modularity amounts to is the claim that the mind can do different things at
different times (rather than the stronger claim that it can do different things at the same
time), and this suffices to call it modular, it ought to be permissible to say that a knife
which cuts both meat and bread is modular. And yet no one thinks of knives as modular
(unless they are Swiss army knives which actually come with different blades). It is I
think instructive that other anatomically nondissociable systems with shared parts, such
as nervous systems, reproductive systems, endocrine systems and the like—all of which
may be singled out for their natural kind properties—are termed “systems.” One never
hears of digestive modules or reproductive modules. The “modules” of developmental
biology and neuroscience which do have shared and reusable elements are an anomaly of
network science. Most biologists, including developmental biologists, continue to think of
modules as “anatomically distinct parts that can evolve independently” (Wolpert 2011, p.
115). Limbs and vertebrae would be modular on this view (being organs), but not the
larger anatomical systems they comprise.

It is well worth stressing here that my argument should not be read as an instance
of mere carping or terminological pedantry. There are certainly occasions when scruples
over the use of words reveal carping tendencies, and nothing much beyond that, but this
is not one of them. Philosophers and cognitive scientists who allege a “module” for this
capacity and a “module” for that capacity must be taken to be saying something
66
substantial, i.e. something more than merely the fact that we have the capacities in
question. To dignify these capacities with the honorific title “module” is, I would
suggest, an attempt to invest the capacities with special-purpose, special-structure status.
If philosophers and cognitive scientists persist in referring to modules for X and Y in the
face of contrary evidence (i.e. evidence suggesting that the X and Y “modules” are not
special-purpose, special-structure devices), they betray a willingness to exploit the
connotations of a powerful term for rhetorical purposes. For if by alleging that there is a
module for X or Y the speaker intends only to say that we can give systematic accounts of
X and Y—where X and Y represent particular foci of the scientific gaze upon the mind—
the speaker is only avowing a belief in the efficacy of the scientific method in the realm of
cognition, which I take no naturalist to deny. In such circumstances it would be better to
drop the term “module” altogether, and settle for a less loaded (and therefore more
honest) term like “capacity,” “faculty,” or “system.”

So while it is true that I am insisting on correct usage, this insistence is not

without justification, and not without consequences should laxity prevail. In some ways
the issues here are analogous to those which have arisen in the philosophy of biology over
the proper use of the word “innate” (see Chapter 6). Neither those who urge
elimination—because the word engenders confusion and fallacies of ambiguity amid a
plethora of conflicting folk-biological intuitions—nor those arguing that a technical
definition can be given, should be seen as engaging in a merely feeble semantic dispute.

4.3 THE BRAIN MODULE

As I have already mentioned several times in passing, neuroscience gets by for the most
part with a very specific notion of modularity to hand. This is not the sense in which
modules are familiar in network science, nor the sense in which they are familiar in most
of psychology and cognitive science. The neuroscientific module is sometimes called a
“brain module” or “cortical module” (Mountcastle 1978; 1997; Pascual-Leone &
Hamilton 2001; Gold & Roskies 2008; Rowland & Moser 2014; Zador 2015), other times a
“cortical column” or “columnar module” (Mountcastle 1978; 1997; Buxhoeveden &
Casanova 2002; Amaral & Strick 2013; Zador 2015), still at other times an “elementary
processing unit” (Kandel & Hudspeth 2013), or simply an “operator” (Pascual-Leone &

67
Hamilton 2001; Pascual-Leone et al. 2005). As I foreshadowed earlier, it corresponds
roughly with the node of a neural coactivation graph. Slight variations in the meanings of
these terms will not be important in the present context. It is true that the cortical
“column” forms part of a distinctive hypothesis in neuroscience that arguably
contemplates a narrower class of phenomena than is conveyed by the nodes of a network
graph. But nothing need turn on this here. Indeed from one point of view the metamodal
(reusable) node is a fully generalized account of the more specific columnar module
(Jacobs 1999, pp. 33-34; Pascual-Leone & Hamilton 2001, pp. 427-428, 441, 443).

Various formulations of the criteria for modularity have been proposed in

neuroscience (Buxhoeveden & Casanova 2002, p. 940). The general notion is of a
coherent functional unit with a more or less dedicated input-output specification,
somewhat on a par with the modern microprocessor chip (Leise 1990, p. 1). Gazzaniga
(1989, p. 947) assumes “a high degree of functional specificity in the information
transmitted over neural systems,” and that modular organization consists of “identifiable
component processes that participate in the generation of a cognitive state. The effects of
isolating entire modular systems or of disconnecting the component parts can be
observed” (my emphasis). Leise (1990) defines a module as a group of cells with similar
response properties (see also Amaral & Strick 2013, p. 348; Zador 2015, p. 44). Krubitzer
(1995, p. 412) defines them as “structural and physiological discontinuities within the
limits of a classically defined cortical field...reflected in architectonic
appearance…neural-response properties, stimulus preference and connections.” The idea
here is clearly predicated upon both functional and anatomical specificity.

The brain module’s explanatory rationale is simple. As Gazzaniga (1989, p. 947)

concludes from a review of the comparative evidence, “research on animals has led to the
belief that there are anatomic modules involved in information processing of all kinds and
that they work in parallel and are distributed throughout the brain.” In the same vein,
Kandel & Hudspeth (2013, p. 17) state that neuroscientists “now think that all [high-
level] cognitive abilities result from the interaction of many processing mechanisms
distributed in several regions of the brain. Specific brain regions are not responsible for
specific mental faculties” (my emphasis). High-level gross functions such as language,
perception, affect, thought, movement and memory “are all made possible by the

68
interlinkage of serial and parallel processing in discrete brain regions, each with specific
functions” (Kandel & Hudspeth 2013, p. 17, my emphasis; Bressler 1995; Gazzaniga
1989, p. 947). High-level mental functions fractionate into low-level subfunctions, then,
and it is these narrowly defined low-level operating systems that are understood to satisfy
the criteria for modularity in neuroscience. The key principle here is that of distributed
parallel processing, in which “functional parts…interconnect uniquely to form processing
networks” (Krubitzer 1995, p. 408; Bressler 1995; Mountcastle 1997, p. 717). Kandel and
Hudspeth give a vivid illustration:

Simple introspection suggests that we store each piece of our knowledge as a

single representation that can be recalled by memory-jogging stimuli or even by
the imagination alone. Everything you know about your grandmother, for
example, seems to be stored in one complete representation that is equally
accessible whether you see her in person, hear her voice, or simply think about
her. Our experience, however, is not a faithful guide to how knowledge is stored
in memory. Knowledge about grandmother is not stored as a single representation
but rather is subdivided into distinct categories and stored separately. One region
of the brain stores information about the invariant physical features that trigger
your visual recognition of her. Information about changeable aspects of her face—
her expression and lip movements that relate to social communication—is stored
in another region. The ability to recognize her voice is mediated in yet another
region. (2013, pp. 17-18)

This picture fits flush with the sort of distributed parallel activation evidence that
underpins neural reuse (Pasqualotto 2016; Pessoa 2016). Indeed, to the extent that they
are not strictly domain-specific, the stable low-level operations that occur as nodes in
these distributed systems seem to be the empirical equivalent of the low-level cognitive
workings posited in the earliest formulations of the massive redeployment hypothesis.

A little history will clarify the significance of this discovery. The elaboration of
the distributed processing model is the high point of an intense research effort within the
structuralist tradition. In my earlier discussion I noted that Carl Wernicke stood out
among the ranks of modern neurologists with his distinctive vision of the structure-
function relation. I suggested that he may even have been operating with an implicit
understanding of the difference between a cognitive working and a cognitive use
(Bergeron 2007). In a famous paper, Wernicke (1908) described a novel kind of aphasia,
one in which the patient can produce words but not comprehend them—the precise
inverse of the pathology described by Broca earlier that century. The brain lesion
69
responsible for this aphasia was to a distinct cortical region of the left cerebral hemisphere
(later called Wernicke’s area). Wernicke presented his account of this pathology in terms
of an explicit neural model of language processing that attempted to steer a middle course
between the two competing frameworks of his day, that of the phrenologists and cellular
connectionists on the one hand, who contended that specific functions were realized in
localized neural tissue (and were therefore guided by the anatomical modularity
assumption), and that of the holists on the other hand, who supposed that every mental
function involved the brain as an aggregate (Kandel & Hudspeth 2013). Wernicke’s
model had only basic sensory-motor and perceptual functions localized to discrete regions
of cortex. Higher functions depended on the cooperation of several neural elements,
implying that single behaviours could not be pinned down to specific sites. Wernicke thus
became the first neurologist to advance the thoroughly modern notion of distributed
processing (Kandel & Hudspeth 2013; Mountcastle 1997). He assigned a specific
language motor program governing the mouth movements for speech to the region
implicated in Broca’s aphasia, and a sensory program governing word perception to the
area implicated in the new aphasia he described.

According to this model, the initial steps in neural processing of spoken or written
words occur in separate sensory areas of the cortex specialized for auditory or
visual information. This information is then conveyed to a cortical association
area, the angular gyrus, specialized for processing both auditory and visual
information. Here, according to Wernicke, spoken or written words are
transformed into a neural sensory code shared by both speech and writing. This
representation is conveyed to Wernicke’s area, where it is recognized as language
and associated with meaning. It is also relayed to Broca’s area, which contains the
rules, or grammar, for transforming the sensory representation into a motor
representation that can be realized as spoken or written language. When this
transformation from sensory to motor representation cannot take place, the
patient loses the ability to speak and write. (Kandel & Hudspeth 2013, p. 12)

The success of Wernicke’s clinical model in predicting a third type of aphasia—one in

which “the receptive and expressive zones for speech are intact, but the neuronal fibers
that connect them are destroyed”—as well as its general influence among late nineteenth
century neurologists, helped inaugurate a new approach to cortical localization
spearheaded by the German anatomist Korbinian Brodmann. Brodmann’s revolutionary
method of distinguishing cortical regions on the basis of cellular shape and vertical

70
orientation brings us one step closer to the cortical columns that are now taken to be the
“fundamental computational modules of the neocortex” (Amaral & Strick 2013, p. 348).

Brodmann’s contribution was to extend the histological and cytoarchitectonic

methods of his day by working comparatively, i.e. across species. He showed that neurons
in the cerebral cortex have both a layerwise (laminar) and vertical (columnar) orientation
(Fig. 3(a)), and used this structure to guide his subdivision of the brain into more
functionally discrete regions. Specifically, Brodmann noted differences in the packing
densities and shapes of neurons as he bore down into the cortex, as well as differences in
laminar thickness and synaptic connections as he traveled horizontally along its surface.
This proved to be a decisive step, for we now know that functional differences in cortex
depend on the relative thickness of layers as one moves from region to region. Each of its
six layers is characterized by different inputs and outputs, with neurons projecting to
different parts of the brain. “Projections to other parts of the neocortex, the so-called
cortico-cortical or associational connections, arise primarily from neurons in layers II and
III. Projections to subcortical regions arise mainly from layers V and VI” (Amaral &
Strick 2013, p. 346). “Input” areas such as the primary visual cortex receive sensory
information from the thalamus, and therefore have an enlarged layer IV since this is
where axons from the thalamus typically terminate: “The input layer contains a
specialized type of excitatory neuron called the stellate cell, which has a dense bushy
dendrite that is relatively localized, and seems particularly good at collecting the local
axonal input to this layer” (O’Reilly et al. 2012, p. 33). “Hidden” areas, processing
neither inputs or outputs but essential to the formation of abstract category
representations, are thickest at layers II and III, since the predominance of pyramidal
cells in these layers makes them “well positioned for performing this critical
categorization function” (O’Reilly et al. 2012, p. 34). Finally “output” areas have their
thickest layers at layers V and VI, given that the efferent connections which typify output
zones must “synapse directly onto [subcortical] muscle control areas,” and it is the
neurons in these layers which best meet this requirement (O’Reilly et al. 2012, p. 34).
Brodmann marked the boundaries where these surface differences occurred and was thus
able to distinguish the 47 distinct brain regions that have since become synonymous with
his name (Fig. 3(b)). Each of Brodmann’s brain areas consequently relates to a specific

71
cognitive or sensory-motor function: areas 44 and 45, for instance, correspond to Broca’s
area, and area 22 corresponds to Wernicke’s area.

(a) (b)

Figure 3. (a) A cross-section of the cortical sheet revealing its layered organization in accordance with cell
type. (b) Brodmann’s famous brain map. Broca’s area is in Brodmann areas 44 and 45, close to primary
motor cortex (Brodmann area 4). Wernicke’s area is in Brodmann area 22, close to primary auditory cortex
(Brodmann areas 41 and 42).

This is where modules re-enter the story. The sort of cytoarchitectonic methods
which Brodmann employed, while delivering a very useful functional subdivision by the
standards of his day, were not quite able to do justice to the subtlety of functional
variation in the cortex. For the five regions Brodmann designated as being concerned
with visual function (areas 17-21), modern electrophysiological and connectional analyses
have interposed 35. These take the form of cortical columns that run from the outermost
surface of the cortical sheet (the so-called pial surface) to the white matter deep beneath
layer VI. A column is in effect a very thin cross-sectional slice of the cortical field, no
more than a fraction of a millimetre across, such that “[n]eurons within a cloumn tend to
have very similar response properties, presumably because they form a local processing
network” (Amaral & Strick 2013, p. 348). It is this distinctive columnar structure which
passes for the basic cognitive module of neuroscience today (Mountcastle 1997; Zador
2015), and its importance resides, partly at least, in the computational efficiency it confers
on neural circuits:

Columnar organization…minimizes the distance required for neurons with

similar functional properties to communicate with one another and allows them to
share inputs from discrete pathways that convey information about particular
72
 sensory attributes. This efficient connectivity economizes on the use of brain
volume and maximizes processing speed. The clustering of neurons into
functional groups, as in the columns of the cortex, allows the brain to minimize
the number of neurons required for analyzing different attributes. If all neurons
were tuned for every attribute, the resultant combinatorial explosion would
require a prohibitive number of neurons. (Gilbert 2013, p. 570)

At least part of the motivation for the brain module, then, is to address concerns around
the scaling problem we encountered in § 3.2 (i.e. as the number of neurons increases, the
number of neurons that must be connected grows quadratically larger). It is genuinely
modular in the sense of possessing both functional specificity—i.e. a discrete
computational operation definable over a preferred (but nonexclusive) set of inputs—and
spatial localization (Pascual-Leone & Hamilton 2001, pp. 441, 443; Gazzaniga 1989, p.
947; O’Reilly et al. 2012, pp. 36-40; Pasqualotto 2016; Pessoa 2016).

All this is predominantly (and paradigmatically) true of the sensory-motor cortical

maps discussed in Chapter 2. Many of these have “functionally specific, connected
neurons to extract behaviorally relevant features [e.g. lines and edges from spatial
receptive fields] from incoming sensory information” and “a degree of functional
autonomy” (Rowland & Moser 2014, p. 22). Whether this organization is exemplified also
by nonsensory/nonmotor high-end association cortices has not up until now been clear,
but Rowland & Moser (2014) review evidence suggesting that there are definite
similarities between sensorimotor columns and the organization found in medial
entorhinal cortex (MEC) implicated in episodic and spatial memory tasks. If the grid map
of MEC really were to be organized in this modular fashion, it would certainly put paid to
the idea of a rigid Cartesian distinction between “central” and “peripheral” cognition as
far as modularity is concerned (see § 7.2.2). Of course the precise degree to which MEC
resembles columnar organization is the crucial question. The similarities for their part are
clear: MEC has “vertically linked cells, tight bundling of dendrites from the deeper
layers, and predominantly local connections raising the possibility that it contains
functionally autonomous columns” (Rowland & Moser 2014, p. 22). Moreover, “MEC
has well-defined spatial responses that allow the cells to be analyzed for topography and
modularity in their response properties” (Rowland & Moser 2014, p. 22). There is one
noteworthy difference, however. The majority of entorhinal modules appear to be
anatomically intermingled such that while they remain functionally independent and

73
discrete (dissociable in principle) they are anatomically overlapping and spatially
interspersed, rather than strictly localized. Entorhinal modules therefore appear to be
merely functional, not anatomical. Their functional specificity is further corroborated by
the fact that, although columns are themselves composed of far smaller units called
“minicolumns” (consisting of between 80-100 neurons), “[n]o research has yet
determined the capacity of minicolumns for independent activity outside the
macrocolumn that they belong to” (Buxhoeveden & Casanova 2002, p. 937). The upshot
of all this is that the brain could be organized into column-based modules of roughly
common form throughout, including regions that are important to central cognition.

What needs most emphasizing about the brain module are those very qualities
which set it apart from the classical notion that still features unmistakably in discussions
of modularity within cognitive science, cognitive neuropsychology, neuropsychology and
the philosophy of mind. Here I am referring to its extremely restricted scope—an
exiguously small subfunctional computation—and its dynamic metamodal response
properties: the brain module is in essence a domain-general reusable operator appearing
within various interacting, nested and distributed neural assemblies (Mountcastle 1997;
Jacobs 1999; Pascual-Leone & Hamilton 2001, Pascual-Leone et al. 2005; Pasqualotto
2016; Pessoa 2016). We saw these dynamic response properties in connection with an
earlier discussion revolving around crossmodal plasticity, supramodal organization and
domain specificity (§§ 2.4.2-2.4.3). I shall revisit and elaborate on this material in the next
section, when I explain more fully the character and import of Pascual-Leone &
Hamilton’s (2001) original metamodal hypothesis of brain organization. It will be relevant
both on the issue of the functional specificity of modules (§ 5.1) and their early
development (Chapter 6).

Thus far I have provided an outline of the varieties of modularity, defended what
I take to be indispensable in any modular theory of the mind, and foregrounded the
neuroscientific notion of modularity. The next chapter pursues head-on the implications
of neural reuse for the modularity of mind.

74
 4.4 SUMMARY

In recent decades neuroscience has challenged the orthodox account of the modular
mind. As I have shown, one way of meeting this challenge has been to go for increasingly
“soft” versions of modularity, and one version in particular, which I dub the “system”
view, is so soft that it promises to meet practically any challenge neuroscience can throw
at it. But an account of the mind which tells us that the mind can do different things,
even interesting things, is not itself necessarily an interesting account. In this chapter I
have reconsidered afresh what we ought to regard as the sine qua non of modularity, and
offered a few arguments against the view that an insipid “system” module could be the
legitimate successor of the traditional notion. In part my arguments can be read as a plea
for the precise use of language, but there is more than pettifogging behind this plea.

75
 5
_____

Modules reconsidered: Whither modularity?

5.1 DOES MODULARITY SURVIVE THE EVIDENCE OF NEURAL REUSE?

One of the primary contentions of this chapter is that the cortical column we have just
examined is probably the only robust example of modularity that could survive evidence
of reuse, and this just because reuse seems almost destined to predict something very
much like it: small, stable, reusable nodes appearing within various distributed networks
spanning various cognitive domains. The question before us now is whether despite
appearances neural redeployment really is compatible with the degree of functional
specificity that modularity demands.

One thing appears reasonably certain. If the cortical column (or Andersonian
working) were to survive reuse as the dedicated and functionally specific cognitive unit
that it would need to be, not only would reuse then be compatible with the modularity of
mind, it seems fair to say the Fodorian module itself would be likely to survive in some
form—at least to the extent that cortical columns retain both stimulus specificity and
informational autonomy, properties which they are likely to retain if brain regions are as
task-selective and functionally constrained as the evidence in § 2.4.3 suggests they are. To
be sure, the neo-Fodorian module would be a shadow of its former self, barely
recognizable in point of size, and certainly no longer suited to its original role as a marker
of high-level cognitive function.1 But the resulting picture of the mind would still be

1 By describing Fodor’s modules as “high-level” I mean only that they can be specified at the level of

proprietary domains (e.g. at the level of vision, olfaction and language rather than simply at the level of

76
modular, and Fodor did after all contend for the modularity of mind (albeit in a very
distinctive sense).

But does the cortical column—or its Andersonian concomitant—emerge

unscathed in this way? To get a sharper sense of the options available to us on this
question, I shall set the overall account of reuse in the context of Pascual-Leone and
Hamilton’s (2001) original metamodal hypothesis, which is an important forerunner of
contemporary theories of reuse, including Anderson’s. This account trades in brain
modules, which it terms “operators,” and so allows me to convey very crisply the obvious
sense in which modularity is compatible with reuse. I shall then walk through the
principal objections to this view. Given what I take modules to be, my criterion of
demarcation must be the degree to which dissociability no longer remains tenable even in
principle. If functional specificity is no more than a will-o’-the wisp, modularity itself can
be little more than that. To that end, I shall propose a simple device by which we can
usefully conceptualize the problem facing the modularist. At its core, modularity turns on
evidence of specialization. What we require, therefore, is a scale of specificity for brain
regions which makes their indicia of specificity explicit. As far as I am aware, such indicia
have not been propounded in any detail before. 2 I conclude this section with an
assessment of the long-run prospects of modularity.

The metamodal hypothesis is intended to account for the observation that “our
perceptual experience of the world is richly multimodal”—that “[w]e are able to extract
information derived from one sensory modality and use it in another,” and “know a shape
by touch and identify it correctly by sight” (Pascual-Leone & Hamilton 2001, p. 427).
The hypothesis accommodates the possibility of crossmodal recruitment, and more
specifically, the supramodal dynamics we encountered in § 2.4.3. Of relevance here is the
fact that it is an adaptation of Robert Jacobs’ (1999) “mixtures of experts” (ME) model.
The ME model builds on two important ideas. First is the idea of functional specificity

edge-detection or depth discrimination). They are not high-level in the sense that they pertain to complex
thought, judgment or memory. See § 7.2.2 for comment on Fodor’s central/peripheral distinction.
2 I hasten to add, however, that Anderson’s (2014) “dispositional vector” account of brain regions is an
alternative strategy for coming to grips with the same set of issues. Others are clearly alive to the problem.
Proponents of the Leabra architecture, for instance, resist modularist terminology precisely because it
“forces a binary distinction on what is fundamentally a continuum” (Petrov et al. 2010, p. 287). See also
Frost et al. (2015).

77
and spatial localization (i.e. the anatomical modularity assumption). Different brain
regions possess different structural properties, and these differences make for differences
in functional capability to the extent that some regions will be better suited to performing
particular functions over others. Second is the idea of competition between modules.
Brain regions become specialized for processing particular inputs through open
competition, but the competition is rigged, as it were, by the functional proficiencies
which characterize the different regions: “each region tends to win the competition for
those functions for which its structure makes it particularly well suited” (Jacobs 1999, p.
32). Two predictions follow. One is that the differences between neural regions appear
quite early in development, and might even be innate (an issue to which I return in
Chapter 6). The other is that “neural modules should enforce the outcome of a
competition through a set of adaptable inhibitory interactions that allow modules to
suppress the outputs of other modules” (Jacobs 1999, p. 34). Accordingly, Pascual-Leone
and Hamilton propose that, instead of “unimodal sensory systems that are eventually
integrated in multimodal association cortical regions,” the whole cortex

might actually represent a metamodal structure organized as operators that

execute a given function or computation regardless of sensory input modality.
Such operators might have a predilection for a given sensory input based on its
relative suitability for the assigned computation. Such predilection might lead to
operator-specific selective reinforcement of certain sensory inputs, eventually
generating the impression of a brain structured in parallel segregated systems
processing different sensory signals. In this view, the “visual cortex” is only
“visual” because we have sight, and because the assigned computation of the
striate cortex is best accomplished using retinal, visual information. Similarly, the
“auditory cortex” is only auditory in hearing individuals and only because the
computation performed by the temporal, perisylvian cortex is best implemented
on cochlear, auditory signals. However, in the face of visual deprivation, the
“striate cortex operator” will unmask its tactile and auditory inputs to implement
its assigned computation using the available sensory information. (2001, pp. 427-
428)

The crucial message for us here is that on this picture, despite neural operators being
functionally and computationally constrained, their range of application is not. Neural
operators are intrinsically versatile from the point of view of which inputs they can
process, limited only by the amenability of the inputs to undergo a definite sort of
manipulation. Barrett and Kurzban (2006, pp. 634-635) call something like this formal
domain specificity—a construal of domain specificity where “domain” does not refer to

78
the set of objects or stimuli betokened by a traditional task category, but instead to inputs
defined by reference to the formal processing competence of the system in question.
Formal domain specificity is therefore a syntactic construal of domain generality, and sets
the stage for the pervasiveness of redeployment as a basic strategy of neural design. It
comports with a view of the brain in which its several regions have a manifold of latent
afferent input channels—preexisting connections supplying the critical cortical
infrastructure which makes reuse possible. And while the picture presented in the above
quotations would suggest a certain stability or equilibrium is achieved after a suppression
mechanism ensures the best module wins (so that individual modules get tuned to
particular inputs and not others), as the examples presented in Chapter 2 dramatically
attest, we do not have to wait for “visual deprivation” for this hidden complexity to be
“unmasked,” since it is a normal feature of healthy adult brains to exploit these channels
all the time (e.g. when “seeing” the face of a loved one at the sound of their voice, or tools
at the sound of a hammer, etc.). Hence supramodal organization simply entails neural
reuse. Moreover, the model demonstrates how readily modularity can be combined with
redeployment, inasmuch as the latter naturally presupposes the former.

But now we must finally confront the objections to this “minimodule” view which
reuse seems to entail. We can distinguish two broad lines of attack, one weak, the other
far more serious and potentially fatal. It is well to address the weaker one first. Here the
charge is that minimodules are “compatible with an anemic version of localization that
claims simply that individual brain areas do something, and the same thing, however low-
level, simple, or cognitively uninteresting, whenever they are activated” (Anderson 2007c,
p. 164). Such entities can hardly be controversial, since very few people nowadays regard
the brain as a “disorganized mash” (Prinz 2006). Notice that this is the same objection I
raised earlier against the system view of modularity: “when the notion of modularity is
denatured, it turns into a platitude” (Prinz 2006). And for all their differences, neither
systems nor minimodules pretend to solve the evolutionary debugging problem. So we
seem to have another case of truism dressed up as theory.

In the case of minimodules, however, I think the objection can be overplayed. It is

true that minimodules are incapable of offering a simple way through the debugging
problem, and this might be thought to commend the sort of modules defended by

79
evolutionary psychologists instead. But what would be the use of a theory which resolved
puzzles by ignoring reality? A theory must aim to be both tractable and realistic (Coase
1937). As convenient as it would be for us to suppose that modules are independently
modifiable high-level cognitive components, the evidence of neural reuse suggests that
this is not how the brain is organized. So either the problem itself is real and the evidence
of reuse must be explained away, or the assumption that no non(massively)modular brain
could possibly evolve must be set aside. Surely the latter approach would be the more
sensible. The debugging issue itself is to a large extent a symptom of looking at things the
wrong way. If we accept that on some level evolution has to involve the emergence of
functionally exiguous neural parts, and view the engineering problem as being how
preexisting parts might be combined in novel ways, concerns over debugging become far
less pressing. Minimodules, in any case, are not trivial. The mind could have been (and
indeed has been) modeled in very different ways (think connectionism/PDP, holism,
etc.), and a minimodule hypothesis is quite demonstrably falsifiable (unlike, say, the
system view). Minimodules also support robust predictions (like forward inference) and
theory-building. The truth is that minimodules are as modular as they need to be—
modular enough to solve the very real wiring problems posed by scaling circuits, and
modular enough to rule competing accounts like functional holism and strict localization
out of the question. The trivialization charge is a nonstarter.

What then of the more serious line of attack? Although there are developments of
the argument in several directions, its general thrust is to make a lot of the fact that the
brain implements a network. Anderson (2010, p. 249) frames the issue in these terms:
“Instead of the decompose-and-localize approach to cognitive science that is advocated
and exemplified by most modular accounts of the brain, neural reuse encourages ‘network
thinking.’ ” To recapitulate briefly, all networks share a number of important properties,
properties which make the study of any structure that exhibits a network design far more
tractable than it might otherwise be. Preeminent among these of course are nodes and
edges, but a defining mark of the network approach is its focus on the global structure of
interactions between nodes, rather than the individual nodes themselves. Thus if the
brain is a network, modularity goes awry, for even if we were to concentrate all our
energies upon modules qua nodes (e.g. minimodules), still we would be missing the

80
point—the key to networks lies not in their nodes, but in the structure of their
interactions.

Such is the clear-cut statement of the challenge. Put in this form, however, it just
overstates the case. First, the fact that a team of soccer players exhibits higher order
dynamics in no way obviates the importance of individual players to the game; indeed
their unique talents and skills are what drive the interactions which feature at the level of
abstract topology. Second, and this is well worth remembering through all the hype, one
should be no less judicious in one’s use of a network analogy than in one’s use of any
other:

Although the terms “network” and “connectivity” are widely used when talking
about regional covariation in the human brain, it is important to keep in mind that
no human data at present allow us to make inferences about brain regions forming
networks in the true sense of the word. In particular, under a technical definition,
two brain regions form a network if they are anatomically connected, typically via
monosynaptic projections. In living humans, we rarely, if ever, can say anything
conclusive about anatomical connections among brain regions…[C]ollections of
regions are more appropriately characterized as functional systems. (Fedorenko &
Thompson-Schill 2014, p. 121)

Still, even if we were to moderate the argument in allowing for such

complications, the network challenge would remain. We may for convenience describe
three distinct iterations of the challenge, each more persuasive than the last, which in one
way or another play upon the importance of the network context for understanding local
function (inasmuch as context determines meaning). The thought here is that because
minimodules appear across multiple and functionally diverse neural communities, they
lack the precise degree of specialization required of modules—in view of just how tiny
minimodules are, the more partnerships a given minimodule enters into, the more
abstract its contribution becomes and the dumber, simpler and more generic it will
ultimately be (Klein 2012). Price and Friston (2005, p. 268) use the example of a
forefinger. Its many roles could include piano playing, typing, scratching, pinching and
feeding; yet if we had to designate its overall role, we would have to settle on something
explanatorily inert: “the forefinger can only do one thing—‘bend’ and ‘straighten.’ Its
role in other tasks is…entirely dependent on what the other fingers and thumbs are doing
and what environmental context they are in.” In short, “naming the specific function that

81
a region performs (or even supposing it has some single specific function) involves a kind
of abduction that is inherently underconstrained and uncertain” (Anderson 2014, p. 53).

Another way of framing the issue is in terms of plasticity. The more functionally
versatile and unstable a brain region, the more plastic it must be (other things being
equal). At the limit, ontogenetic plasticity might be so great that even sudden, swift
connection changes to the neural configuration of a given brain region between
alternating task demands would be possible, and functional stasis merely illusory. Up
until now neuroscientists have simply presumed that a network approach can naturally
complement a modular approach—naturally, because from the minimodule perspective
modules are nothing more than the nodes of a coactivation graph; but the plasticity of
neural regions might so undermine their functional specificity that even neuroscientists
will have to give up the pretense that nodes can be modules in the full-blown sense they
almost always take for granted, as when they describe nodes as “functionally specific
brain regions” or “regions that are selectively engaged by a particular mental process”
(Fedorenko & Thompson-Schill 2014, p. 121). In the event that neuroscientists might
still like to refer to nodes as modules—much in the way they conventionally use the term
to describe the communities of nodes in graphs—it would be a case of terminological
convenience trumping theoretical rectitude.

The weakest iteration of the challenge adds little to what has already been said,
but it might note how the preponderance of afferent input pathways sustaining the brain’s
supramodal organization must ever so slightly colour an individual module’s operations as
to rob it of a deep and lasting functional essence. The more functionally promiscuous a
region, joining now with the visual system, now with the language system (say), the more
we can expect the neural context to impinge on the region’s functional capabilities. Brain
regions are by and large fairly homogeneous anyway (Buxhoeveden & Casanova 2002, p.
941). Standard histological preparations and cytoarchitectonic methods often fail to reveal
anatomical differences between neighbouring yet functionally distinct cortical columns.
Thus an important strategy by which the brain generates difference from sameness is
through the formation of different interconnection patterns among neurons and regions,
indeed often among the very same neurons and regions. Input channels therefore cannot
be conceived as merely useful appendages to the lines of script run by distinct neural

82
operators, as they are themselves partly constitutive of the operations performed by them.
Functional promiscuity means we cannot retain a prespecified notion of function for
brain regions considered in isolation of the neural contexts in which they appear.

Now it must be said that when put like this the argument again runs the danger of
just overstating its case. For what it seems to lead to is a variety of holism. Insofar as that
is where this line of thinking is taking us, it should be resisted, for the weight of evidence
does not support holism, classical connectionism/PDP, or anything like it really. With
that proviso in place, the argument is a good one—functionally distinct operators with
functionally distinct input criteria can be observed in the brain, but a moderate pitch to
incorporate the effects of context would not go astray. Let us call these “weak context
effects.” Weak context effects are those which do not compromise a brain region’s ability
to perform a well-defined, functionally specific (albeit domain-general) operation. This is
consistent with how Anderson (2010, p. 295) originally defined a working: “Abstractly, it
is whatever single, relatively simple thing a local neural circuit does for or offers to all of
the functional complexes of which the circuit is a part.”

Evidence for stronger context effects are not hard to find. Let us call them “strong
context effects.” These will constitute the basis for the second and third iteration of the
network challenge, but before advancing any further on this front, I should make one
point clear at the outset: there is something about strong context effects—implying as
they do a much higher degree of plasticity for local circuits than we have encountered so
far (details to follow)—which sits uneasily with aspects of the evidence of massive
redeployment presented in Chapter 3. The problem is that strong context effects are
incompatible with evidence suggesting that the units of redeployment are themselves
relatively fixed in nature (even after allowing for synaptogenesis, etc.). To the extent that
strong context effects obtain, then, the theory of reuse requires amendment. Anderson’s
massive redeployment hypothesis, it will be remembered, predicts that recently evolved
functions should be supported by more widely scattered regions of the brain than older
ones, since it should on the whole prove easier to utilize existing circuits than to devise
special purpose circuitry afresh, “and there is little reason to suppose that the useful
elements will happen to reside in neighbouring brain regions” (Anderson 2010, p. 246).
Not only is the evidence which Anderson cites consistent with this prediction, it could

83
also be taken to imply something more specific about the nature of local circuits, an
implication which Anderson originally had no hesitation in drawing:

If neural circuits could be easily put to almost any use (that is, if small neural
regions were locally poly-functional, as some advocates of connectionist models
suggest), then given the increased metabolic costs of maintaining long-distance
connections, we would expect the circuits implementing functions to remain
relatively localized. That this is not the observed pattern suggests that some
functionally relevant aspect of local circuits is relatively fixed. (Anderson 2010, p.
247)

But while this is one way of interpreting the evidence, a distributed network organization
might be favoured by evolution for rather different reasons. As Bullmore and Sporns
(2012, p. 336-337) point out, one reason why a general principle of parsimonious cost
control might be compromised in favour of far-flung neural circuits has to do with the
resilience that such organization may be presumed to confer. Robustness to adverse
perturbations—“[t]he degree to which the topological properties of a network are resilient
to ‘lesions’ such as the removal of nodes or edges”—could well have more to do with the
distributed structure of recently acquired capacities than the functional fixity of local
circuits. At the very least, the inference that local circuits are not especially plastic again
“involves a kind of abduction that is inherently underconstrained and uncertain.”
Anderson himself appears to have moved on from his earlier commitment to fixed local
workings, but not on account of resilience per se. He has lately been convinced by the
evidence of strong context effects in its own right, and as a result no longer speaks of
fixed local “workings,” preferring instead the less rigid connotations of the term “bias” in
describing the functional proclivities of local brain regions. For Anderson a cortical bias
represents “a set of dispositional tendencies that capture the set of inputs to which the
circuit will respond and govern the form of the resulting output” (2014, p. 15)—an idea
which reconciles a brain region’s versatility and its overall functional durability without at
the same time insinuating “that each circuit does exactly one specific thing” (2014, p. 16).

So what exactly, then, are strong context effects? I think we may usefully divide
them into two broad categories. The first category—motivating the second iteration of
the network challenge—would appear to suggest that small brain regions can assume
radically different network states, and thereby alter their basic electrophysiological
configurations, depending on the requirements of the cognitive system being used. This
84
sort of operational, on-the-fly ontogenetic plasticity of neurocognitive resources
undermines the purported functional fixity of brain regions, and hence the claim that
brain regions can be modular (in the true sense of being functionally specialized). The
second category—motivating the third and final iteration of the network challenge—goes
even further than this by throwing into question the very legitimacy of functional
decomposition as a basic strategy within the cognitive sciences. Here the thought is that
“it is not as if we can identify the one fixed function of an element of the system and then
worry about the effect of interactions later. Rather, the interactions are often precisely
what fix local function” (Anderson 2014, p. 208). This may not at first appear to be saying
much more than what was said in the first instance. In fact its ramifications are deeply
unsettling for the “decompose-and-localize” approach to cognitive science, as I shall
explain more fully in a moment. Let us take these two putative categories of context
effects in turn.3

Evidence of swift, sudden connection changes in networks begins at the single

neuron level. C. elegans has acquired fame as the nematode for which the first neural
network wiring diagram was published. It contains about 300 neurons and up to 7000
synaptic connections, simple yet complex enough to serve as a model of function-
structure dynamics within higher organisms. C. elegans neurons perform “more than one
type of circuit function, including both motor and sensory functions,” and sometimes
perform multiple functions within the same modality (Altun & Hall 2011). Beyond the
straightforward implication here that neural reuse may be evolutionarily conserved, there
are intimations of still more intriguing possibilities. Neuromodulation refers to a family of
context effects in which it is possible for the same neuron to radically change function—
and perform in just the opposite role—in response to changes in the electrophysiological,
chemical and genetic environment. One example is C. elegans’ olfactory neuron,
AWCON, which can apparently signal both attraction and repulsion to the very same
odour depending on its neuromodulatory configuration. Another is the nocioreceptive
ASH neuron, which can direct both sociality and avoidance. But neuromodulation is not
restricted to C. elegans. Similar effects have been documented in both the pond snail and
honeybee, and there are enough instances within vertebrates to suggest that
neuromodulation might be a basic evolutionary strategy for coping with scarce neural

3 I am heavily indebted to Anderson (2014) for the review which follows.

85
resources (Anderson 2014, p. 34). Of course before such results could support more
ambitious inferences regarding human cognition, we would need evidence of large-scale
modulation in more complex organisms. In the simplest organisms, small modifications of
even single synapses can have significant behavioural ramifications. In larger and more
complex organisms this is unlikely to be the case. In fact evidence of such large-scale
effects does exist, even within the human literature. Most suggestive of all is the evidence
Cole et al. (2013) report for “flexible hubs” in the brain which “flexibly and rapidly shift
their brain-wide functional connectivity patterns” in response to changing task demands.
If brain regions really do move into different functional configurations, as distinct from
being redeployed in the same state for different purposes, it would imply that brain
regions can be neither functionally specialized (in the sense of contributing a stable and
predictable operation across their various higher order applications) nor dissociable, both
because their disruption would directly impede the operation of an equivalent system—
the selfsame region considered from the standpoint of any of its alternative network
states—and because it could well prove impossible to identify a segregable unit of neural
tissue that retained a constant form from state to state.

One upshot of this concerns theory-building. Any theory we construct that tokens
a brain region subject to strong context effects will not be able to offer a fully general
explanation of what that region offers to all of its networks. Even those who do not think
contextualism would undermine our ability to construct powerful theories supporting
strong predictions concede that we would nonetheless be in the realm of partial
generalizations (Burnston 2016). Furthermore, part of the appeal of a theory which posits
functionally specific brain regions is that it supports robust inferences: one should in
principle be able to infer which brain region has been engaged simply from knowing what
function is being performed. Strong context effects undermine the robustness of such
inferences.

None of this entails that the brain is equipotential or has an inherently open
texture (as we shall see in Chapter 6). On the contrary, overlaps between the neural
implementations of cognitive tasks are frequently found to involve functional and
semantic inheritance (Mather et al. 2013, pp. 109-110; see also § 3.3), implying that brain
regions have a stable set of causal features that regulates their participation in various

86
networks. This is consistent with the finding that recently acquired skills in the human
lineage, such as reading and writing, have highly uniform neural substrates across both
individuals and cultures (Dehaene 2005). But when the specific point in dispute is
whether the mind/brain has a modular architecture, such facts alone cannot be decisive,
for then the issue is not whether brain regions have specific developmental biases, input
preferences, or an underlying structural and functional integrity, but precisely the degree
to which brain regions are specialized. A bias is not a specialization.

To recapitulate, so far we have considered how natural the alliance between

modularity and reuse can be, and proceeded to examine various objections to this view.
The objections come in two forms, weak and strong. The weak objection alleges that
minimodules are trivial entities, but we saw how this claim is in fact unwarranted. The
stronger objection plays on the network structure of brain organization to reveal the
illusoriness of functional specialization for individual brain regions, being merely network
nodes whose functional importance is subordinate to internodal network interactions.
This stronger network challenge in turn assumes three distinct forms, one emphasizing
weak context effects (which we dismissed as instructive but not fatal to modularity), and
two emphasizing strong context effects. The first category of strong context effects
correlates with increasing ontogenetic plasticity. The objection from these context effects
really does have bite, and probably compromises the modularity of any brain region that
is vulnerable to their impact. I turn now to the second category of strong context effects.

The second category raises very serious doubts over the legitimacy of
componential analysis, and so by implication practically all mainstream work in the
cognitive sciences. The decompose-and-localize approach to cognition assumes that the
mind can be understood on the analogy of a machine with working parts. Central to this
approach is the belief that function can be explained in terms of “bottom-up additive
contributions” rather than “top-down constraints that limit, change, or determine
functional properties” (Anderson 2014, p. 308). Recent discoveries suggest this
confidence may be misplaced, although quite to what extent remains unclear at this stage.
The starkest illustration of these effects is offered by starburst amacrine cells (SAC) in
the mammalian retina. These are axonless neurons with dendrites arranged radially
around their cell body. “What is especially interesting about these cells is that the

87
dendrites are individually differentially sensitive to motion; they release neurotransmitter
only in response to motion that is congruent with the direction the dendrite points away
from the cell body” (Anderson 2014, pp. 92-93). It is tempting to think of each dendrite
as a component because each appears to contribute uniquely and dissociably to effects at a
higher network level.

In fact, the directional selectivity of each dendrite is due in large part to the
particular blend of connections these cells have to bipolar cells and other SACs
such that responses in the congruent dendrites are reinforced while responses in
noncongruent dendrites are inhibited. In other words the directional selectivity of
the dendrite in a given situation is due not so much to intrinsic properties of that
dendrite but to global properties of the network. Global function is not built from
componential local function, but rather the reverse! (Anderson 2014, p. 93)

While one could well think that the entire local network is itself a component, it should
not come as a surprise if the very same dynamics “reproduce themselves at the higher
level,” with the functional selectivity of whatever putative higher level component being
determined again by global network properties rather than intrinsic local features. If these
dynamics apply more generally to neural networks, the assumption behind componential
analysis would be substantially undermined, for then no longer would components be
“temporally and functionally stable subassemblies sitting on the tabletop waiting for final
construction” (Anderson 2014, pp. 93-94). Instead the “functional organization of the
whole” would be logically prior to the functionally parasitic part. Put another way,
interactions between parts would be more important than the activity of parts (Anderson
2014, p. 40).

Olaf Sporns has recently mooted similar ideas. The traditional way of thinking
about circuits is in terms of “highly specific point-to-point interaction among circuit
elements with each link transmitting very specific information, much like an electronic or
logic circuit in a computer” (Sporns 2015, p. 92). On this account, the activity of the
whole circuit is “fully determined by the sum total of these specific interactions,” with
the corollary that “circuit function is fully decomposable…into neat sequences of causes
and effects.” This is a Laplacian model of classical mechanics, “with circuit elements
exerting purely local effects.” The modern approach from complexity theory and network
science, however, emphasizes “that global outcomes are irreducible to simple localized
causes, and that the functioning of the network as a whole transcends the functioning of
88
each of its individual elements.” As an example of an emergent network phenomenon
Sporns takes neural synchronization, “the coordinated firing ‘in sync’ of large numbers of
nerve cells” (Sporns 2015, p. 93). While this phenomenon clearly depends on elemental
interactions and synaptic connections, “it is not attributable to any specific causal chain of
interactions in a circuit model.” Rather, it is “the global outcome of many local events
orchestrated by the network as a whole.”

We can represent these varying degrees of modular specialization along a

continuum running from A to E, each with the indicia represented in Table 1. Brain
regions at or to the left of C, which marks the onset of weak context effects, will be
sufficiently specialized to count as modular. Brain regions to the right of C, characterized
by strong context effects, will not. Plasticity increases as one moves from A through D. A
possible network architecture for types A, B, C and D is shown in Figure 5.

So will modularity survive evidence of neural reuse, neuromodulation, and the

very strongest effects of network context? On the one hand, and from a purely pragmatic
standpoint, I think it would be premature to come down firmly on one side or the other of
this question without first having more solid evidence about the causal properties of
individual brain regions and how they contribute to overall cognitive function. Precisely
how the causal properties of tiny brain regions facilitate more complex functions is not
understood, so the analysis above is really offered as an informed best estimate rather than
as a conclusive demonstration. In addition, it is worth remembering that at this stage the
case for the very strongest of context effects is still speculative. Russell Poldrack, for his
part—whose laboratory work in this space has been pioneering (e.g. Poldrack et al.
2009)—is convinced that cognitive systems will bottom out in low-level, domain-general
and functionally specific computational operations bearing a one-to-one relation to
specific cortical sites. On the other hand, if neuromodulatory and context effects are
indeed as pervasive and game-changing as some people seem to think (e.g. Bach-y-Rita
2004), perhaps only a few scattered islands of modularity are all we can reasonably hope
for (Prinz 2006). It is true that employing current techniques it is not actually possible to
assign brain regions to definite locations on a continuum, so we cannot know for sure that
only a few brain regions will cluster towards the specialist end (e.g. A through C above).
But evidence for the existence, power and ubiquity of context effects can only proliferate

89
at this point, one would think. (Incidentally—taking up a point I raised in § 4.3—if the
“modules” reaching into central cognition turn out to have type D characteristics, central
cognition will be pro tanto nonmodular after all. See § 7.2.2 for further discussion.)

Increasing plasticity

A B C D E

Theoretical Strict Formal Neuro- Non-

domain domain domain modulation decomposition
specificity specificity specificity

Indicia Minimal Few Many Many ⎯⎯

afferent afferent afferent afferent
connections connections connections connections

Participation Participation in Participation Participation ⎯⎯

in a single various tasks & in various in various
task & composites tasks & tasks &
composite within a single composites composites
within a task category within within various
single task various task task categories
category categories

Nonreuse Nonreuse Reuse Reuse ⎯⎯

No context Negligible Weak Dynamic local Local function

effects context effects context network states fixed by global
effects properties

Functional Functional Functional Functional Functional

specialization specialization specialization differentiation differentiation
Example Probably Neural element Extrastriate Flexible Starburst
none—a common & body area; “hubs” in the amacrine cells;
theoretical exclusive to Broca’s area brain reported synchronization
postulate only reading, writing by Cole et al.
& speaking, (2013)
e.g. the
neural basis of
subjacency/wh-
movement (?)

Table 1. A scale of specificity along with indicia of specificity for brain regions.

90
Figure 5. Possible network architecture for brain regions having the indicia of types A through D on the
scale of specificity. Black circles indicate communication with other regions. Source: Petrov et al. 2010, p.
286.

It is worth mentioning that the picture here is consistent with the emerging
consensus around the neocortical column we met in § 4.3 (Rockland 2010; see da Costa &
Martin 2010 for a historical review). Rockland takes five defining features of the column
and argues that these are too rigid to do justice to the complexity of cortical organization.
For instance, it is supposed that columns are solid structures, but this is not quite true,
since they have a heterogeneous substructure that “correlates with reports of locally
heterogeneous response properties,” very much as reuse predicts (Rockland 2010, p. 3).
Their anatomy is therefore messy rather than solid. Columns also form part of widely
distributed networks at several levels (again much as reuse predicts), and for that matter
are not even obligatory to cortex: for instance “comparative anatomy provides many

91
examples of cortex apparently without anatomical columns or dramatically modified
columns,” e.g. whales and dolphins, whose insular cortices have cellular modules
concentrated in layer II, and the giraffe whose occipital cortex has modules concentrated
in the same layer (Rockland 2010, p. 7).

I think it is fair to say, then, that while most of the cortex undoubtedly consists of
module-like elements, probably only a few of these will in the end prove to be modular in
the robust sense we require. The full implications of network thinking for componential
analysis in particular have not sunk in, even though they promise to overturn our
conception of local function almost completely. It goes without saying, of course, that to
the extent that modules do exist, it will be as functionally exiguous and promiscuous
network nodes. The days of classical modularity are well and truly over.

5.2 CAN COMPOSITE SYSTEMS BE DISSOCIABLE?

Up to this point in the discussion I have simply assumed that a cognitive system
consisting of shared domain-general parts cannot be separately modifiable ipso facto.
Some, however, have maintained that neural overlaps need not undermine the functional
independence of high-level cognitive functions (Carruthers 2006, pp. 23-24). This is just
to raise the possibility that high-level cognitive functions could persist as functional
modules (as distinct from anatomical modules). For instance, it is undoubtedly true that
of any two cognitive systems considered in isolation the extent of neural overlap may be
only partial (or even negligible) (Fig. 6a). This would render the two systems dissociable
vis-à-vis each other to the extent that a modification not affecting shared components
would disrupt or improve the affected system independently. Moreover

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. (Carruthers 2010, p. 289)

The plausibility of this suggestion diminishes the more one takes the message of neural
reuse to heart. The evidence of reuse suggests that neural overlaps are a pervasive feature
of cognitive systems, so the chances of true functional independence are going to be quite
low (Fig. 6b). Taking high-level cognitive systems two at a time looks to be a futile
92
strategy. Once the true scale of reuse dawns, the claim that high-level cognitive functions
may persist as functional modules appears far less convincing. And to be clear, my own
wager is that reuse will be so pervasive that not only neural components but the various
connections between them too will be extensively reused throughout the brain.

(a) (b)

Figure 6. (a) Partial overlap between the neural implementations of two cognitive systems. (b) Pervasive
overlaps between the implementations of various cognitive systems.

5.3 MODULAR NOTATION

Perhaps the single most important upshot of the discussion so far has been that
modularity can no longer serve in the role of marking a traditional high-level cognitive
ontology. We have seen how modules (really “minimodules”) are both structurally and
functionally exiguous, and so nowhere up to the job of supporting functions as complex
as language taken by themselves. In this section I shall provide a simple notation to
express at a glance the essential features of the new perspective I am advocating. It will
serve as a convenient shorthand with which to convey some of the more important points
relating to the search for a language module in Chapter 7. Thus let us take modules to be
defined by the set

{M1, M2, M3 … Mn}

93
Modules are really small networks of neurons, and can for convenience be labelled “M-
networks” (to distinguish them from the many higher level networks in which modules
participate in turn). Thus a module can be defined by the set

{N1, N2, N3 … Nn}

where N denotes a neuron, so that a given module Ma will comprise a set of neurons

Ma : {Na, Nb, Nc, …}

An M-network is (or resembles) the structure which neuroscience variously terms a

“module,” “column,” or “elementary processing unit,” and which Bergeron (2007; 2008),
and Anderson (2010; 2015) originally, called a “working.” It consists of around 6000
neurons or 60-80 minicolumns, each minicolumn consisting of between 80-100 neurons
(Buxhoeveden & Casanova 2002, p. 935). The higher level functional composites in which
modules participate are themselves networks (call them “C-networks”). We can take C-
networks to be defined by the set

{C1, C2, C3 … Cn}

so that a given C-network Ca will be a set of M-networks

Ca : {Ma, Mb, Mc, …}

I take M-networks and C-networks to be the central explananda of cognitive

neuroscience. Given the rather drab prognosis with which I concluded § 5.1, we should
expect to find only a smattering of real M-networks in the cortex, and that many of the
structures which neuroscientists have identified as modules have been technically
misdescribed.

Over the years the attention of psychologists and cognitive scientists has quite
understandably been lavished upon the functional taxonomies which C-networks serve to
implement. But, as I have suggested several times already, the same scientists often

94
thought they were dealing with something having the structural characteristics of an M-
network. This was most unfortunate, and its ill-effects have by no means been eradicated.
Such misconceptions necessarily inform both the design and interpretation of scientific
experiments. The debate about whether the fusiform gyrus is specialized for faces, to take
only one example, “has unfolded in the context of the shared belief that the ventral visual
areas are specialized for recognizing some classes of objects,” a belief which is no longer
tenable pending further notice (Gold & Roskies 2008, p. 354).

To sum up, then, there are at least two networks of interest so far as the
modularity of mind is concerned: the network of neurons which constitutes a node/M-
network, and the network of nodes which constitutes a composite of nodes/C-network.
The former is what has come to be regarded as a module in mainstream neuroscience,
while the latter is regarded as a module among those working with graphs in network
neuroscience. This latter notion, as we saw, has obvious affinities with the mental
modules familiar to cognitive scientists, psychologists and philosophers, since it seems to
track quite readily the ontologies of traditional psychology (language, vision, face-
recognition, etc.). One cannot, however, assume that mental modules (Fodorian or
otherwise) reduce smoothly to the communities of nodes that are studied extensively in
graph theory (Anderson 2010, p. 303; 2014, p. 42). Quite apart from other differences, the
classic Fodorian module is an anatomical module, and hence functionally dissociable and
localized in relatively segregated neural tissue. This is not the case of graph-theoretic
(network neuroscience) modules, as I explained earlier. If Fodor’s module has any
legitimate successor at all, then, it must be something with relative stimulus specificity
and informational autonomy—something with the functional characteristics of an M-
network.

5.4 SUMMARY

In Chapter 4 I argued that we ought to regard dissociability as the sine qua non of
modularity. As for what in the brain actually meets this standard, the only likely
candidate will be something resembling a cortical column. But this is not guaranteed. The
effects of the neural network context may so compromise a region’s ability to maintain a
set of stable input-output functions that it cannot be considered a genuine module.

95
 6
_____

Are modules innate?

6.1 PRELIMINARY REMARKS

Asking whether modules are innate is problematic for three reasons. First, the argument
and conclusion of the previous chapter suggests that the question is not a good one to
begin with: we would do better to ask whether brain regions are innate. Second, there is
the rather thorny issue of what one actually means by “innate.” Third, the question
assumes that a general answer can be given, when it is unlikely that all (or even most)
brain structures will have the same developmental story to tell: “there will be cases and
cases” (Mameli & Papineau 2006, p. 564). This last concern can be alleviated by
concentrating the weight of one’s empirical attention on a particular module, if not by
having regard to as much brain-wide evidence as possible. Chapter 2 is my attempt to
incorporate a wide survey of the evidence of neuroplasticity—with as many caveats and
limiting clauses as its interpretation reasonably warrants—while Chapter 7 is my attempt
to home in on one particular system (namely language). In the present chapter I aim to
build on the interpretation of neuroplasticity that I began in Chapter 2.

This still of course leaves us with the problem of having to define what we mean
by innateness, a far from trivial matter (Griffiths 2002; Mameli & Bateson 2006; Bateson
& Mameli 2007; Mameli & Bateson 2011). The trouble is that the term is as ambiguous as
it is entrenched, and some have wondered whether it can perform a useful function in the
sciences at all. With so large a variety of distinct notions lying beneath the surface, it
becomes very easy to commit fallacies of ambiguity (Griffiths et al. 2009). One might for
example infer that a trait is species typical in virtue of its being the product of natural
selection, or developmentally fixed in virtue of its being species typical (O’Neill 2015). In
the result Griffiths (2002) recommends having done with the term altogether, and
suggests that scientists should specify explicitly what they mean on any given occasion. I
do not take an eliminativist stance myself, but, along with Griffiths, do think it absolutely
essential to make explicit the sense in which the term is being used. Following O’Neill
(2015), what I have in mind is insensitivity relative to some specific set of environmental
variations. This is the idea of developmental robustness or environmental canalization
broadly speaking (Ariew 1996; 1999; 2007; Mameli & Bateson 2006; O’Neill 2015), except
that it is explicitly relativized to specific environmental factors (Bateson & Mameli 2007,
p. 823).1 After all, no trait is developmentally robust in an absolute sense, and yet the
claim is frequently made without specifying the environmental factors with respect to
which the trait is supposed to be robust. For the most part this is not a problem, since it is
usually clear in a given context which environmental factors are relevant (O’Neill 2015, p.
212). Still it is important to bear in mind that a trait’s invariance (or otherwise) is always
relative. In the present context, we are concerned with the innateness of modules—
iterated cortical structures with distinctive columnar and laminar patterns of
organization. It should by now be clear that modules are not insensitive with respect to
such experiences as learning, injury and sensory deprivation, regardless of how young or
mature the organism happens to be. The extent of both intramodal and crossmodal
plasticity, as well as evidence for the extensive rewiring of latent supramodal connection
channels, does much to discredit the traditional nativist assumption of “hardwired”
cognitive capacities with rigid developmental schedules (Marcus 2004).

And yet this cannot be the full story. For one thing, sensitivity with respect to a
particular set of environmental factors does not entail sensitivity with respect to others;
and in the absence of factors to which a trait is sensitive, its development might well be
considered robust. For instance, when developmental biologists speak of “activity-
independent” cell differentiation, which results in cortical areas acquiring fixed structural

1 I say broadly speaking because strictly speaking canalization results in a “buffered” developmental
pathway in which insensitivity with respect to some environmental factor is the result of a specific
mechanism or evolutionary adaptation geared to that end (e.g. Waddington 1953; 1955). But insensitivity
simpliciter can be the result of an environmental factor’s having no causal influence on a trait at all. A fly’s
wing pattern could be insensitive to certain pesticides without having been buffered against them by natural
selection, e.g. because the pesticides concerned do not interact causally with the fly’s development in any
way (O’Neill 2015).

97
characteristics in utero, they can be taken to imply that at least some aspects of modular
development are insensitive relative to certain factors, although obviously not with
respect to those factors which can be expected to become influential at a later stage of
development, i.e. during postnatal “activity-dependent” cell differentiation (Saitoe &
Tully 2001, p. 193; Kolb et al. 2001, p. 225; Sanes & Jessell 2013, p. 1259).2 Furthermore,
sensitivity admits of degrees (Collins 2005). Granted that cortical development is robust
in certain respects, how robust is an important question in each case. Thus there are
really two senses in which we can speak of invariance as being a matter of degree: along
one dimension, we can say that the more factors with respect to which a trait is robust,
the more invariant it will be; while along a second dimension, the more a trait is robust to
variation in any single factor, the more invariant it will be (Griffiths & Machery 2008, p.
399). One may therefore legitimately inquire as to whether cell differentiation results in a
stereotyped but essentially crude pattern of synaptic connections and brain regions before
birth, or whether it results in more robust operations that limit and constrain the
functions these regions can later take on. There is scope for genuine disagreement here
between those who think there is a lot of prewiring, combined with some inevitable
rewiring during development (Marcus 2004), and those who think there is comparatively
little prewiring, with a lot of rewiring during development and later life (see the
discussion by Mameli & Papineau 2006, pp. 563-564).

In this chapter I shall argue that the evidence of neuroplasticity supports neither a
traditional nativist nor yet strictly antinativist interpretation of development. Rather, we
seem to be confronting a phenomenon that falls somewhere midway between the two
extremes of developmental hardwiring and original equipotentiality. While the extent of
the neuroplastic responses we considered in Chapter 2 is undoubtedly impressive, and
sometimes vast, a closer look at these cases suggests that the pattern of responses is
constrained. For all their plasticity, brain modules and regions appear to be significantly
robust in the presence of such environmental variables as learning, injury and sensory
deprivation. More precisely, the changes that do occur are exactly what one would expect
to find on the assumption that cortical regions have robust processing capabilities and
clear input preferences (what I earlier described as a “bias”). This is not a traditional

2Moreover, a trait’s sensitivity with respect to a set of experiences at one stage of development does not
preclude its being insensitive with respect to the same experiences at an earlier stage (Kolb et al. 2001, pp.
223, 225; Mameli & Bateson 2006, p. 169; see also § 2.3).

98
nativist picture, to be sure, but neither is it antinativist. (Warning: the evidence to follow
is circumstantial, the argumentation nondemonstrative and abductive. In the realm of
cognition, however, we frequently find ourselves with little else to go on.)

6.2 IMPLICATIONS OF NEUROPLASTICITY

Of all the instances of cortical map plasticity we reviewed in Chapter 2, undoubtedly the
most impressive involve crossmodal changes in which brain regions deprived of their
typical inputs come to subserve alternative uses. One example I mentioned there
concerned early blind Braille readers whose visual cortex appears to be functionally
important for Braille character identification, suggesting a functional contribution of the
reorganized occipital cortices during complex tactile discrimination tasks (Sadato et al.
1996). Moreover, when repetitive transcranial magnetic stimulation (rTMS) is used to
impair the functioning of the occipital cortex, blind subjects appear to have difficulty
performing embossed character recognition, while sighted control subjects do not, again
pointing up the functional significance of early blind occipital cortices during tactile
discrimination (Cohen et al. 1997). Probably the most famous case of crossmodal
plasticity is that of the rewired ferrets whose visual cortex was induced to project into
auditory cortex after their retinal nerves were rerouted so that instead of feeding into
primary visual cortex, they fed into primary auditory cortex via the auditory thalamus
(Sharma et al. 2000; Melchner et al. 2000). The manipulation resulted in ferret auditory
cortex taking on features typical of occipital cortex, such as columnar orientation and
stimulus selectivity. Besides these cases, language studies suggest that this sort of
plasticity is not confined to sensory-motor cortices alone, as the case of EB discussed in
Chapter 2 illustrates very well.

While these results seem quite dramatic, nevertheless some aspects of the
evidence do not sit well with the idea of the brain as open-endedly malleable. In fact,
rather than supporting the case for plasticity tout court, these results argue the case for
what Laurence and Margolis (2015) call “constrained plasticity.” Take the ferret case.
The clear suggestion here is that auditory cortex came to resemble the processing
structures typically associated with occipital cortex. And indeed to some extent this is
what seems to have happened. But in fact primary occipital cortex is a complicated

99
structure, “connected to a large number of distinct brain regions that support further
specific types of visual processing, including computations responsible for downstream
representations of location, direction of motion, speed, shape, and so on” (Laurence &
Margolis 2015, p. 127). And there is no evidence that any of this complex processing
structure was reproduced, for “the overall wiring of the ferrets’ auditory cortex was
largely unchanged.” One interpretation of why a “largely unchanged” auditory cortex was
able to process visual inputs is consonant with the theory of supramodal organization (and
Pascual-Leone and Hamilton’s metamodal hypothesis). Recall that this theory posits a
large number of intrinsically stable neural operators that are more or less suited to
processing specific types of input, but which are at the same time metamodal in that they
receive inputs from many domains (i.e. they are really domain-general, or formally
domain-specific). From this perspective, we would naturally expect there to be something
about visual and auditory stimuli that makes them ideal for a neural operator whose
processing disposition makes it suited to process one or the other of these specific types of
input. Bregman and Pinker (1978) long ago postulated high-level analogies in
computations which involve auditory and visual stimuli (e.g. different pitches are
analogous to different locations, pronounced changes of pitch are analogous to sudden
changes in the direction of motion, etc.). If such analogies between hearing and vision
hold, it would suggest—consistently with the metamodal hypothesis—that auditory
cortex did not really need to change when it began to receive inputs from a domain to
which its processing capabilities were already well suited. As Laurence and Margolis
interpret the ferret case:

even though the rewiring experiments show that the auditory cortex can be
recruited for a certain amount of visual processing, this is because the auditory
cortex and the visual cortex overlap in the types of computations they naturally
support. Far from being a model case of the environment instructing an
equipotential cortex, [the ferret] rewiring experiments illustrate the way in which
cortical structure and function remain largely unchanged even in the extreme case
of input coming from a different sensory system. (2015, p. 128)

Next consider the case of EB from Chapter 2. EB recovered most of his language skills
two years after undergoing a left hemispherectomy at the age of two and a half and tested
as largely normal with respect to language at age fourteen, albeit with his language faculty
now subserved by regions in his right cerebral hemisphere. Surely this argues for an
almost equipotential cortex early in development, if anything does? Not quite. The fMRI
100
evidence shows that the pattern of activation in his right hemisphere is almost isomorphic
to that of the left hemisphere in normal control subjects, revealing a definite and
predictable cortical pattern. Language did not arbitrarily migrate to a new location: it
moved to the very site in the right hemisphere whose structural features most nearly
resemble those of the left hemisphere’s language circuits. A truly equipotential brain
would presumably reconfigure cortical sites selected on a far more ad hoc basis. The most
important take-home message here, then, is not that the brain is open-endedly plastic, but
rather that “the brain’s two hemispheres incorporate a large measure of potential
redundancy of function that can be exploited at certain stages of development” (Laurence
& Margolis 2015, p. 126; see also Jungé & Dennett 2010, p. 278; Barrett & Kurzban 2006,
pp. 638-639; § 7.5).

These cases are only a beginning. By far the most significant evidence for
constrained plasticity and the robust development of brain regions comes from studies
revealing the brain’s latent supramodal organization. A flavour of this evidence was given
in § 2.4.3, but it is instructive to consider a few more examples to drive the point home. It
will be remembered that evidence of supramodal organization first came from studies of
the two major visual processing streams, i.e. the dorsal (“where”) path for space and
motion discrimination, and the ventral (“what”) path for object and shape recognition.
What these studies suggest is that this dual stream processing structure persists with the
same functional role and structural characteristics in both early and congenitally blind
and sighted subjects. That is to say even total and protracted visual input inhibition—
from the very earliest developmental stages onwards—appears to have few if any adverse
effects on the development of typical visual processing structures in humans. To repeat
the conclusion one researcher drew from the case we examined in Chapter 2, “despite the
vast plasticity of the cortex to process other sensory inputs” these findings suggest
“retention of functional specialization in this same region” (Striem-Amit & Amedi 2014,
p. 4). The dorsal and ventral processing streams appear to be modular, developmentally
constrained and functionally preserved despite complete early and congenital visual
impairment.

In one study (Renier et al. 2010), early blind subjects were presented with paired
auditory stimuli that differed either in type (in this case, different piano chords) or

101
locality.3 The task required subjects to indicate whether the pairings were of the same
type or emanated from the same location. Subjects exhibited differential activation in a
region of the dorsal visual stream—specifically, the area rostral to the right middle
occipital gyrus (MOG)—when engaged in the auditory spatial location task relative to the
sound-type identification task. Similar results were obtained on an analogous tactile
discrimination task using the same subjects. So while the MOG is clearly plastic, in that
early blind individuals recruit this area more intensively for auditory and tactile
discrimination tasks than sighted individuals, its plasticity reveals it to be functionally
constrained and structurally preserved. It is classically supramodal in that it continues to
perform a fixed computation despite receiving different sensory input. Other studies
attest to the persistence of the spatial location function of the dorsal visual stream.
Consider the posterior parietal cortex (PPC), implicated in the spatial representations that
guide action. In healthy sighted subjects, caudal subregions play a relatively larger role in
reaching and grasping than rostral subregions, which are primarily engaged in the
planning and execution of action. Lingnau et al. (2014) showed that the same response
gradient occurs in the congenitally blind, concluding that “neural plasticity acts within a
relatively rigid framework of predetermined functional specialization” (2014, p. 547).
Other studies evidence preservation of the direction representation function of the dorsal
visual stream as judged by performance of congenitally blind subjects on analogous
auditory discrimination tasks (Wolbers et al. 2011), as well as functions in the ventral
visual stream in both congenitally blind and blindfolded sighted subjects (as we saw in
Chapter 2) (Striem-Amit et al. 2012; Striem-Amit & Amedi 2014). Laurence and
Margolis conclude their review of this evidence in the following way:

it would appear that the large-scale functional architecture of the visual cortex—
the division of labor between the dorsal and ventral streams—develops in much
the same way, and with the same functions being performed in various subregions
of these streams, with or without visual experience. (2015, p. 133)

And of course all of this evidence once again testifies to the supramodal organization of
the brain, and Pascual-Leone and Hamilton’s metamodal hypothesis in particular, since it
is consistent with a brain that is composed of a number of “distinct computational
systems whose functions are established independently of their sensory input” (Laurence

3 The examples in this paragraph and the next are drawn from Laurence and Margolis (2015).

102
& Margolis, p. 428) and in which “multimodal sensory inputs feed into all cortical
regions” (Pascual-Leone & Hamilton 2001, p. 432), even though the operations of a given
region will dictate certain preferences. The metamodal hypothesis predicts that “when
the preferred input is unavailable, the brain switches to the next best fit” (Laurence &
Margolis, p. 428) such that a region’s underlying computational structure and profile
need undergo no truly radical alteration in the face of new processing inputs—in the
standard case it will perform in much the same way it always did, albeit on a new set of
afferents. On this view, even many dramatic instances of crossmodal plasticity, where the
equipotential nature of the cortex seems to be its most obvious feature, need involve little
more than a straightforward remodeling of supramodal connection channels (Pascual-
Leone & Hamilton 2001, p. 443).

One final study is especially worth mentioning for the illumination it provides on
the precise extent to which predefined cortical functionality is developmentally robust. A
group of mice whose brains were genetically modified so that they were incapable of
synaptic transmission, and therefore incapable of releasing any neurotransmitters at all,
were compared to normal control littermates. Mice in whom the potential for all synaptic
transmission has been inhibited in this way have effectively no potential for learning or
indeed any activity-dependent cell differentiation. Verhage et al. (2000) reported that, at
least prior to birth, the two brain types were assembled correctly, and were in fact
essentially similar. As they state their own findings:

Neuronal proliferation, migration and differentiation into specific brain areas

were unaffected. At [embryonic day 12], brains from null mutant and control
littermates were morphologically indistinguishable….At birth, late-forming brain
areas such as the neocortex appeared identical in null mutant and control
littermates, including a distinctive segregation of neurons into cortical
layers….Furthermore, fiber pathways were targeted correctly in null mutants.
(2000, p. 866)

This means activity-independent changes are robust enough to withstand severe synaptic
privation, and “that many features of even the fine-grained structure of the brain can
develop without any sensory input or feedback” (Laurence & Margolis 2015, p. 130).

Notice, incidentally, just what this sort of neuroconstructivist nativism implies:

that while there is a certain (and relative) sense in which M-networks and other
103
functionally significant brain regions are innate or robust, the same cannot be said for
high-level cognitive functions. There is a weak sense, of course, in which the innateness
of M-networks translates to the innateness of high-level functional composites, which are
innate insofar as the parts used in assembling them are innate. But this claim is different
from the claim that such higher level composites are innate as organized (Jusczyk &
Cohen 1985). If the “derived” innateness of a functional composite were sufficient for its
being considered innate as an organized ensemble, all complex cognitive functions would
be innate by default, which is plainly absurd. I shall revisit this matter in Chapter 6.

Before I leave this chapter, it will be useful delineating once again the relationship
between neuroplasticity (qua Hebbian learning) and neural reuse, for there is a good deal
of complementarity on offer here that is easy to miss amid the detail of specific cases. The
supramodally organized brain in effect constitutes the architectural foundation upon
which Hebbian synaptic mechanisms operate. That is to say, Hebbian plasticity
presupposes reuse, inasmuch as it consists in the strengthening (or weakening) of existing
supramodal connection channels. Synaptic pruning, synaptogenesis and other forms of
interneural transmission can no doubt account for the more drastic examples of plastic
change and postpathological recovery we examined in Chapter 2 (yielding “a change in
use from a change in working,” in the language of Chapter 3), perhaps joining a suite of
mechanisms that could account for the very youngest cortico-cortical pathways
established in the developing brain (in effect supplying us with a supramodal
architectural foundation).4 But Hebbian mechanisms remain an important part of the
story of how patterns of neural reuse are regularly refined and remodeled in the course of
normal development, learning and recovery after injury (yielding “a change in use
without a change in working,” as we saw in Chapter 3).

6.3 SUMMARY

The brain’s plasticity is definitely constrained. While plasticity is an intrinsic and crucial
feature of the nervous system, it is important to emphasize that the brain is not open-
endedly plastic. Furthermore a brain region can be innate in a relatively strong sense and

4Anderson (2014) hypothesizes such a suite of mechanisms under the label “search.” I return to this idea in
Chapter 7, with a twist of my own.

104
yet fail to reach the threshold characteristics of a genuine module. A bias, after all, is not a
specialization.

105
 7
_____

The language module reconsidered

7.1 PRELIMINARY REMARKS

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 honours 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 vigourously 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,

106
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 thesis 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 within theories of cognitive architecture than most proponents of neural
reuse—and indeed cognitive scientists generally—seem willing to do. It strikes me that in
view of how simple and powerful the principle is, it is a wonder that more has not been
said about it. To my mind it is a significant omission, although happily one which 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 below, 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

107
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 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 to 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 framework. 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 redundancy has received scant attention
in the philosophical and cognitive neuroscience literature. The basic idea here is that, no
doubt for good evolutionary reasons, the brain incorporates a large measure of
redundancy of function (I suggested as much during my discussion of neuroplasticity: see
§ 6.2). 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 the one hand and fast/automatic or “peripheral” functions on the other.
This is puzzling because the degree of cognitive impenetrability involved plausibly calls
for segregated circuitry (the encapsulation problem). Redundancy naturally explains the

1 The term “solitarity” is a neologism. See § 7.5.

108
data here too. In fact for all we know redundancy might help to 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 biobehavioural 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 near
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.

7.2 DEFINING A LANGUAGE MODULE

7.2.1 The meaning of linguistic specialization

There is a clear consensus in modern neuroscience that language is mediated by “defined

sets of circuits” (Fisher 2015, pp. 150-151). The main debate over these circuits concerns
whether they are specific to language (Chomsky 2005; 2010). In Chapter 3 I raised the
possibility that, despite extensive evidence of the reuse of neural circuits and what
appears to be the deeply interpenetrative nature of mental functions, some small
component or set of components is rarely coopted outside the language domain. Such a

109
component (or set thereof) would be strictly specialized for language in its being recruited
predominantly, perhaps even exclusively, for linguistic purposes. By way of example I
mentioned the possibility of a neuron or restricted set of neurons being dedicated to
conjugating the verb “to be” and having no nonlinguistic functions at all (other examples
are discussed in § 7.2.3 below). I said that this component might aptly be described as a
language “module.” The debate over the specialization of linguistic function, then, can be
understood as a debate concerning the existence of such modules (Fedorenko &
Thompson-Schill 2014). It is an important question in its own right, of course (cf. Fitch
2010), but carries further implications for other inquiries into the human cognitive
system, as well as for the evolution of language. Among the various alternative ways of
construing the issue (as to which see below), this is the understanding with which I shall
proceed here. Let me, however, define the problem more precisely before I turn to
address it directly in the following sections of this chapter.

So far we have seen how the evidence of neural reuse strongly suggests that the
only dissociable unit we are likely to encounter in the brain will be one that resembles the
neuroscientific notion of a module. The neuroscientific module is sometimes called a
“brain module” or “cortical module” (Mountcastle 1978; 1997; Pascual-Leone &
Hamilton 2001; Gold & Roskies 2008; Rowland & Moser 2014; Zador 2015), other times a
“cortical column” or “columnar module” (Mountcastle 1978; 1997; Buxhoeveden &
Casanova 2002; Amaral & Strick 2013; Zador 2015), still at other times an “elementary
processing unit” (Kandel & Hudspeth 2013), or simply an “operator” (Pascual-Leone &
Hamilton 2001; Pascual-Leone et al. 2005). It corresponds roughly with the node of a
neural coactivation graph, and is known to perform only exiguous subfunctions such as
aspects of edge detection or depth discrimination—certainly nothing as high-order as
language acquisition or norm acquisition per se. High order complex functions are instead
enabled by neural ensembles or composites, which are just so many arrangements of these
low-level neural modules, often highly distributed across the cortex (and so not localized,
contrary to much traditional speculation). But we also saw that, owing to the effects of the
many different neural contexts in which modules appear (namely the functional
assemblies instantiating high-level complex functions), it is not clear that such units will
always possess the requisite degree of specialization required to sustain their modularity:
in many cases the label “module” may actually be a misnomer. The true extent of

110
modularity in the cortex—even with the benefit of a neuroscientifically informed
conception to hand—is very much an open question. As a way of getting to grips with
this issue, in Chapter 5 I provided a scale of specificity for brain regions which makes
their indicia of specificity explicit. I reproduce the features of these brain regions below
for convenience. Situating the question of the modularity of language within this
framework sharpens the issue considerably and shows up useful points of contrast with
alternative construals. Varying degrees of modular specialization can be represented along
a continuum running from A to E, each with the indicia as specified below (Table 1).
Brain regions at or to the left of C, which marks the onset of weak context effects, will be
sufficiently specialized to count as modular. Brain regions to the right of C, characterized
by strong context effects, will not. Again, plasticity increases as one moves from A
through D.

The search for a language module may be construed as a search for a type B
module. Let us call such a type B language module an “elementary linguistic unit,” or
“ELU.” It will also be remembered that in § 5.3 I provided a notation to describe the
entities in view here. A true module (any of types A through C) is a certain sort of
network of neurons, which I called an “M-network” (for convenience we may regard all
of the types A through E as M-networks even though the paradigm cases encompass only
A through C). A “C-network” is the composite structure which brings the several
modules implicated in a high-level complex function into coalition. Language as a high-
level psychological capacity is mediated by a restricted class of C-networks (e.g. a speech
comprehension network, a speech production network, etc.). This much is beyond
dispute. The proponent of a language module needs to show in addition that at least one
of these language C-networks’ constituents is an ELU. Indeed the traditional claim is
more ambitious, with theorists maintaining that there is in effect a large M-network that
handles core aspects of language—a super-sized ELU, as it were—such as Chomsky’s
Merge or Fodor’s sentence parser (see below) (Chomsky 1980a, pp. 39, 44; 1988, p. 159;
2002, pp. 84-86; Fodor 1983; Plaut 1995; Pinker & Jackendoff 2005, p. 207; Fitch et al.
2005, p. 182; Collins 2008, p. 155; Fedorenko & Thompson-Schill 2014). The argument
of the present chapter is that there are unlikely to be any ELU’s—that the only units we
are likely to find among the constituents of our language C-networks are M-networks of
the types C through E.

111
 Increasing plasticity

A B C D E

Theoretical Strict Formal Neuro- Non-

domain domain domain modulation decomposition
specificity specificity specificity

Indicia Minimal Few Many Many ⎯⎯

afferent afferent afferent afferent
connections connections connections connections

Participation Participation in Participation Participation ⎯⎯

in a single various tasks & in various in various
task & composites tasks & tasks &
composite within a single composites composites
within a task category within within various
single task various task task categories
category categories

Nonreuse Nonreuse Reuse Reuse ⎯⎯

No context Negligible Weak Dynamic local Local function

effects context effects context network states fixed by global
effects properties

Functional Functional Functional Functional Functional

specialization specialization specialization differentiation differentiation
Example Probably Neural element Extrastriate Flexible Starburst
none—a common & body area; “hubs” in the amacrine cells;
theoretical exclusive to Broca’s area brain reported synchronization
postulate only reading, writing by Cole et al.
& speaking, (2013)
e.g. the
neural basis of
subjacency/wh-
movement (?)

Table 1. A scale of specificity along with indicia of specificity for brain regions.

Now it may seem that this construal of the matter is austere, and that I have set a
most demanding test for the modularity of language. Other ways of understanding
linguistic modularization have occasionally been discussed. The neurolinguists Evelina
Fedorenko and Sharon Thompson-Schill (2014), for example, outline three, the first two

112
of which I shall mention here. The first is the more or less conventional one I have just
described which looks for an ELU. As they put it, “a network may be functionally
specialized for mental process x if all of its nodes are functionally specialized for x, [but]
perhaps the presence of at least one functionally specialized node is sufficient to qualify
the whole network as functionally specialized” (2014, p. 121). On this view even a single
ELU would suffice as evidence of the specialization of language. Obviously I have no
argument with this approach. The second approach, on the other hand, would count as
specialized any system whose pattern of interconnections between nodes is unique to the
function the system performs:

In this approach, the properties of the nodes are less important; they may be
functionally specialized, domain general, or a mixture of the two. What matters is
whether a unique combination of nodes and edges is recruited for the relevant
mental process x. If so, such a network would be considered functionally
specialized for x, even if all of the individual nodes are domain general…and even
the same exact combination of nodes can contribute differently to different mental
processes when the nodes are characterized by different patterns of connection.
(2014, p. 121)

On this much more liberal view, language is specialized if the patterns of connections that
characterize its C-networks are unique to those networks, notwithstanding that the same
(indeed even the very same) nodes are recruited beyond the language domain, provided
that the wiring patterns are distinctive in each case. Now I should think no one would
deny the importance of network configurations when explaining cognitive function, or
that there are occasions when our attention is properly captured by the dynamics of
distinct (yes, specialized) networks; but it would surely surprise no one apart from a holist
that the brain enters into a different state whenever it switches between tasks. Systems
specialized in this sense lack the stability and permanence that provide the sort of
specialization likely to be of interest to those in search of a language module. What has
predominantly mattered to these researchers is just the extent to which mental processes
like language rely on dedicated mechanisms and specific computations.

John Collins, for instance (a philosopher and noted defender of generative

linguistics), conjectures that “the peculiar specificity of language deficits suggests that the
realization of language is found in dedicated circuitry, as opposed to more general levels
of organization” (Collins 2008, p. 155). Chomsky himself has written that “It would be

113
surprising indeed if we were to find that the principles governing [linguistic] phenomena
are operative in other cognitive systems….[T]here is good reason to suppose that the
functioning of the language faculty is guided by special principles specific to this domain”
(Chomsky 1980a, p. 44). Barely a decade later he wrote that “[i]t would be astonishing if
we were to discover that the constituent elements of the language faculty enter crucially
in other domains” (Chomsky 1988, p. 159). Many commentators (e.g. Goldberg 2003;
Pinker & Jackendoff 2005) frequently assume that Chomsky has relented in his stridency
concerning this requirement, but in fact he has continued to hold out for the potential
vindication of “earlier versions of generative grammar” in this regard (see e.g. Fitch et al.
2005, p. 182 and the ambivalent remarks in Chomsky 2010, p. 53; Berwick & Chomsky
2016, p. 91; Chomsky, personal communication). Besides, despite the abstractness of the
Minimalist Program—which simplifies the idealization to language in the interests of
evolutionary tractability—Chomsky has continued to write of a “language organ” that is
“analogous to the heart or the visual system or the system of motor coordination and
planning,” commenting approvingly of the view which regards specialized learning
mechanisms as “organs within the brain” that are “neural circuits whose structure enables
them to perform one particular kind of computation” (Chomsky 2002, pp. 84-86). Pinker
and Jackendoff (2005, p. 207) also defend something like this, pointing to neuroimaging
and brain damage studies suggesting that “partly distinct sets of brain areas subserve
speech and non-speech sounds,” evidence that speech perception “dissociates in a
number of ways from the perception of auditory events.”

For this reason I have construed the issue of linguistic specialization along
traditional lines. I turn next to the other aspect of the problem of defining a language
module.

7.2.2 The domain of language clarified

In one sense defining the language domain ought to be a simple affair, for is it not just
that domain which encompasses activities such as speaking and signing, and (on a broader
plane) reading and writing? The straightforward answer to this is yes, but the complete
picture is somewhat more complicated by the deep and really rather mysterious
relationship between thought and language. It is clear that language expresses a speaker’s

114
thoughts, and that whatever many other purposes a language may serve it always comes
down to the ability to convert sound (or some other signal) into meanings, and meanings
into sound (Chomsky 1980b, p. 46; Sterelny 2006, p. 24; Jackendoff 2007, p. 2;
Christiansen & Chater 2016, pp. 114-15). From this perspective it is natural to view
language as serving some sort of coding function, and the language faculty as a cognitive
system that enables translation between mentalese and strings of symbols (Pinker 1994, p.
60). On such a view there would seem to be at least two (potentially overlapping but
functionally distinct) interacting systems of interest: a thought or “central” system on the
one hand, and a coding or translation system on the other.2 One system generates and
processes thoughts, the other encodes and decodes them. The second system takes its
input from the first during production tasks, while the first takes its input from the
second during comprehension tasks. This is admittedly crude and schematic; there are
also many who would question the aptness of a conduit metaphor for language (Evans &
Levinson 2009, pp. 435-436; Smit 2014). Nonetheless I think the picture is reasonable. As
Justin Leiber (2006, pp. 30-31) puts it, the “commonplace distinction that psychologists
and linguists use [takes] speaking and hearing to be ‘encoding’ and ‘decoding’—i.e.,
converting thoughts, or mental items, into the physical speech stream, and converting the
physical speech stream into thoughts, or mental items.” Certainly a more useful analogy
in the present context would be hard to find, since disputants in the debate over linguistic
modularity can be roughly grouped in accordance with how broadly they construe the
language domain—as we shall see, there are those who would have it encompass (or even
reduce to) thought, and those who would restrict it to the coding function alone.

Chomsky’s (1965; 1975; 1979; 1980a; 1995; 2002; 2005; 2010; 2016) many
iterations of the language module have one thing in common in their portrayal of a central
system that encompasses the very mechanisms of thought (McGilvray 2014, p. 59;
Collins 2004, p. 518). In a collaborative paper, Hauser, Chomsky and Fitch (2002)
distinguished between the faculty of language in a narrow sense (FLN) from the faculty
of language in a broad sense (FLB). FLN as a subset of the mechanisms underlying FLB
is “the abstract linguistic computational system alone, independent of the other systems
with which it interacts and interfaces” (Hauser et al. 2002, p. 1571). Their assumption is

2Within the framework I have been pursuing here, these systems would be construed as two distinct (if
possibly overlapping) C-networks.

115
that “a key component of FLN is a computational system (narrow syntax) that generates
internal representations and maps them into the sensory-motor interface by the
phonological system, and into the conceptual-intentional interface by the (formal)
semantic system” (Hauser et al. 2002, p. 1571). Furthermore “a core property of FLN is
recursion,” which yields discrete infinity and is suggested to be the only uniquely human
and uniquely linguistic cognitive possession (Hauser et al. 2002, p. 1571). The property of
discrete infinity allows the generation of a limitless array of hierarchically structured
expressions from a finite base of elements—the same property which (it is alleged)
generates the system of natural numbers (Chomsky 2005; 2010). The technical term for
this operation is Merge, which in its simplest terms is just set formation (Berwick &
Chomsky 2016, pp. 10, 98). Merge combines words (“Lexical Items”) and sets of words,
taking their semantic information (“features”) to a semantic interface (SEM—the
“conceptual-intentional system”) and their sound information to a phonetic interface
(PHON—the “sensory- motor system”). Merge is therefore a system that generates
sentences (“expressions”) in an inner symbolic code or language of thought (an “I-
language”) (Chomsky 2005, pp. 3, 4; 2010, pp. 55, 59).

It is important to be clear about what conception of language lies behind this

proposal. It is easy to be misled by talk of a phonetic interface, the mappings to that
interface and indeed the whole sensory-motor apparatus, which along with the semantic
system is supposed to be a system for linking sound and meaning. This tends to imply
that the production of an acoustic signal for the purpose of externalization and
communication is what language is for. But this is actually only “the traditional
assumption” (Chomsky 2010, p. 54). The “primary relation” of interest is supposed to be
that between the core faculty of language (FLN) and SEM, i.e. the “systems of thought”
(Chomsky 2010, pp. 54-55). Expressions that satisfy the interface conditions of SEM
yield a “language of thought,” and it is hypothesized that “the earliest stage of language,”
which supposedly arose prior to externalization, was “just that: a language of thought,
available for use internally” (Chomsky 2010, p. 55). This inner code was the unique
possession of a privileged individual, Prometheus, whose language provided him with
“capacities for complex thought, planning, interpretation, and so on....[which] would
then be transmitted to offspring, coming to predominate” (Chomsky 2010, p. 59). It is
easy to forget that because externalization and communication came later, the language of

116
Prometheus was not just a silent inner speech, as the residue of an internalized
conventional public symbol system might be. Rather it is something like the reflexively
complex but wordless stream of thought available to (presumably) any member of Homo
sapiens not yet exposed to a public language.3 For language is “virtually synonomous with
symbolic thought” (Chomsky 2010, p. 59, quoting Ian Tattersall), and “fundamentally a
system of thought” (Berwick & Chomsky 2016, p. 102). Perhaps the clearest indication
that for Chomsky language is the acme of central cognition are recent remarks suggesting
that language functions as a means of integrating information from various proprietary
domains: “…language is the lingua franca that binds together the different
representations from geometric and nongeometric ‘modules,’ just as an ‘inner mental tool’
should. Being able to integrate a variety of perceptual cues and reason about
them…would seem to have definite selective advantages” (Berwick & Chomsky, pp. 165-
166). This makes Prometheus’ language a “language of thought” in pretty much the
classical sense (Fodor 1975). Thus when Chomsky implores us to consider how difficult it
is not to talk to ourselves, both during sleep and our almost every waking hour (Berwick
& Chomsky 2016, p. 64), to press the point that language is really an instrument of
thought, it is important not to assume (no matter how reasonably) that he is extolling the
virtues of a public language. The powerful scaffolding which a public language provides
in the form of an echo for our ideas and ruminations—the chance to objectify and
insinuate our thoughts into a manipulable format external to ourselves, surely what makes
language able to serve as a “tool for thought” par excellence—cannot be denied, of
course, and Chomsky certainly does not (e.g. Berwick & Chomsky, p. 102). But his
primary aim here is not to make the case for externalization so much as to point up the
intimate and virtually indissoluble relation between a Promethean private language and
internal thought. For language here ultimately means something other than what most
people, and I suspect what most language researchers, think about when they think about
language (see below). Most researchers would understand the coding function to be a
distinct system for the translation of thought into the sentences of a public language, even
if this system can be decomposed into elements that are shared with other systems
(including systems of thought). Now just what all this implies for an ELU we shall come
to presently, but first let me contrast Chomsky’s view with Jerry Fodor’s, who seems to

3 By “wordless” I mean without the words of a public language.

117
have a more conventional—Chomsky would say “traditional”—understanding of what I
have called the coding function.

Fodor has consistently maintained that only peripheral input systems are likely to
be modular. On this view modules are associated with specific channels of sensory
transduction—there may be modules for vision, olfaction and hearing, but likely not for
complex thought, memory and judgment. I have two points to make about this, the first
somewhat ancillary to the second. In light of what I have discussed in previous chapters,
this way of construing the difference between central and peripheral systems seems
definitely mistaken. The material I presented in Chapters 2 through 5 demonstrates that
elements of even our most evolutionarily ancient transduction systems participate in
various cross-domain functional composites (C-networks), including those underlying
central processes. Transduction dynamics, which are usually characterized by a certain
degree of speed, autonomy or reflexivity, may even be activated in many cases by the
same domain-general nodes (M-networks/modules) which yield central system
dynamics. This might in fact explain the frequent penetrability of perception. Now as I
have been at pains to show, there is a case to be made for the modularity of mind. But it
fails to get a grip where Fodor would like. And besides, even if there were dedicated
sensory-motor systems, it would still be unlikely that a sharp central/peripheral
distinction at the modular level could be maintained. For Fodor the distinction is sharp,
but this is because for him it holds by fiat: he has in effect simply stipulated that a module
is a device for the processing of transduced information. The possibility that cognition
might be underwritten by anatomically or functionally exiguous units throughout—the
basic assumption in cognitive neuroscience, as we saw—is not actually excluded by
Fodor: it is just that he has construed the term “module” to mean something quite
specific, and that something does not extend to the autonomous columns that handle low-
level subfunctions right across the neocortex (Mountcastle 1978), long understood to be
the seat of complex thought and executive function. But Fodor does not own the term,
and the modular hypothesis—under that very name and always referring to the
functionally specialized units of the mind/brain—goes back at least to the 1950s,
appearing in works by Vernon Mountcastle (1957; 1978), David Marr (1976) and Noam

118
Chomsky (1980a) well before the appearance of Fodor’s (1983) monograph.4 As Collins
(2004, p. 506) summarizes the Fodorian attitude to the central systems: “for Fodor,
whether there are ‘central’ modules is at best moot; the thesis that it’s all modules he
considers to be virtually a priori false.”

The point about Fodorian modularity I want to impress, however, is not that I
think it draws a distinction that is arbitrary so far as the modularity of mind is concerned
(it may be aptly drawn at some other level of inquiry, e.g. an evolutionary one); it is that
his understanding of modularity leads directly to a certain kind of language module, one
very different from Chomsky’s (Collins 2004). Since modules for him are peripheral
input devices, it follows that any language module must be peripheral, and thus not the
sort of system which generates expressions in an inner symbolic code, as Chomsky’s does.
Fodor’s language module is a “sentence encoding-decoding system”—a parser, with an
encapsulated representation of grammar (Fodor et al. 1974, p. 370). Language is for him
“a psychological mechanism that can be plausibly thought of as functioning to provide
information about the distal environment in a format appropriate for central processing”
(Fodor 1983, p. 44). On this account language is not a central process, not pure symbolic
thought, as it is for Chomsky; rather it is a “psychological mechanism” that provides grist
for the central system mill (i.e. for the inner “language of thought”).

All this can make for confusion in debates about the modularity of language. It is
not hard to see how interlocutors might talk past one another. Does a mechanism
recruited exclusively for thought, or perhaps for thought and a more peripheral coding
operation—but nowhere else across cognition—count as an ELU? Or must the
mechanism be exclusive to the coding operation alone before it can be considered an
ELU? It depends on whether you view systems of thought as forming part of the domain
of language. Evidently some do and others do not. Take metarepresentation as a case in
point, the capacity for nested thinking which allows us to embed thoughts within
thoughts, in principle indefinitely, witnessed in a child’s being able to draw a picture of
themselves drawing a picture (Suddendorf 2013; Zerilli 2014). If it could be shown that
metarepresentation is an exclusive property of thought, or an exclusive property of

4 I do accept, of course, that Fodor performed a tremendous service in shaping the discussion of modularity
in cognitive psychology.

119
thought and the coding function taken together, metarepresentation would count as an
ELU on a Chomskian interpretation of language (defined in terms of thought). For
someone with a more traditional understanding, by contrast, metarepresentation would
not count as an ELU, for though it might appear in the coding function, it is exploited
outside the language domain (defined in terms of processes that operate distinctly from
thought), in this case within the systems of thought.

Morten Christiansen and Nick Chater are two psycholinguists who appear to have
the more traditional understanding of the language domain in mind. Among the various
factors they cite to explain why natural languages appear to be so well suited to the
human brain, and hence easy to learn and process, they include “constraints from
thought” (Christiansen & Chater 2016, p. 51). This form of explanation makes most sense
from the point of view that language and thought are not synonomous (otherwise the
explanation would be uninformative). It is just as well, then, that Christiansen and Chater
indeed do regard “constraints from thought” as “non-linguistic constraints” (2016, p. 50).
While I shall adopt this more traditional construal of the language domain, in § 7.3 I
survey evidence of the extensive reuse of language circuits across domains having nothing
much to do with either language or thought. In other words the material I present below
should be problematic for anyone defending the existence of an ELU, regardless of how
eccentrically they wish to construe the language domain.

7.2.3 Examples of elementary linguistic units

Before leaving this section, I should provide some further guidance on the most likely
candidates for the role of an ELU. Now that we have clarified both in what respects an
ELU would be specialized and in what sense it could be linguistic, we can turn to some
concrete proposals.

Much of the impetus for the claim that the mind/brain contains ELU’s came
from early work in generative linguistics, which formalized a large stock of highly
intricate and apparently system-specific rules for the derivation of grammatical strings
(“surface structures”) from the more abstract “kernel” sentences (“deep structures”)
underlying them (Chomsky 1956; 1957). These unspoken deep structures were

120
hypothesized to be “present to the mind” whenever a speaker produces the surface forms
of her language (Chomsky 2006, p. 16). This inspired the belief that the mind/brain
incorporates specialized systems which function more or less exclusively for the
generation of surface structures. While the field of generative linguistics today would
hardly be recognizable to an undergraduate familiar with work from (say) the mid-late
1960s, the influence of that early work has not dissipated entirely, and it is, for all that
times have changed, still plausible to suppose that at least some linguistic operations are
domain-specific. Let me illustrate with a simple example drawn from the generative
tradition.

The assignment of phonetic interpretations to surface structures might hint at

cognitive resources which, in virtue of how detailed and context-specific they seem, could
reasonably be supposed to serve no other function. Assume that a speaker has
encountered the following phonetic realizations:

expedite à expeditious
contrite à contrition
ignite à ignition

Assume further that the speaker has not yet encountered the word “righteous,” so has not
yet been in a position to establish the derivation

right à righteous

The speaker on hearing “righteous” (properly so as “rahy-chuh-s”) for the first time
knows that the underlying form cannot be the same as for expeditious, contrition, and so
on (unless the case is just an exception), though had the speaker heard “rish-uh-s” they
would not have hesitated in concluding that “rite” would be the underlying form
(analogously to expedite/expeditious, etc.). The speaker understands that the underlying
form of “righteous” must instead be “right” (or, more technically, a form containing “i”
followed by the velar continuant “gh”), for only some such form could make sense of
what was heard given the following rule (which the speaker must be taken to know):

“t” followed by a high front vowel [“-eou,” “-iou,” “-ion,” “-ian,” etc.] is realized
as “ch” [as in chew, choke, challenge, etc.] after a continuant [e.g. “–ahy,” as in
121
 fight, bight, sight, etc.—as opposed to “i” as in fit, bit, sit, etc.], and as “sh” [as in
shoe, show, sham, etc.] elsewhere.

Detailed phonological rules of this kind—in fact much more intricate ones than this—
have frequently been thought to reflect principles not obviously assimilable to other
cognitive domains, pertaining exclusively to the coding function. This accompanies the
thought that such rules are so exotic as far as the agent’s overall envelope of capacities go
that handling them must require a very special suite of neural and computational
resources.

Pinker and Jackendoff (2005) suggest other rules. They observe that many
grammatical principles have no real application outside language, principles such as linear
word order (John loves Mary/Mary loves John), agreement (the boy runs vs the boy run),
case (John saw him vs John saw he), tense, aspect, derivational morphology (run/runner)
and inflectional morphology (the girls are/the girl is). Moreover, they contend that
linguistic recursivity is not reducible to analogues in mathematics. They also nominate
speech perception as possibly uniquely adapted for the perception of human speech
sounds (and not other types of sounds). Brattico and Liikkanen (2009, p. 261), in passing,
suggest that the lexicon, as “a list of feature bundles,” is domain-specific. Their argument
is in fact that the only truly domain-specific aspects of language will turn out to be
nongenerative—generative mechanisms (recursion/Merge) will be domain-general.

Actually the question of what it takes to be domain-specific, or “specialized for

X,” can be a little more complicated. For instance, associative learning is the paradigm
domain-general cognitive capacity. But a particular learned association, say between fire
and warmth, could well be considered domain-specific. The specific associative
mechanism linking fire and warmth may be discretely localized in the brain and active
only in response to those specific stimuli. Similarly, we might have a general capacity to
run recursive algorithms, but a particular implementation of that procedure, say a
numerical one, may be domain-specific. It might be that we use the same token procedure
to recognize faces and words. But it is perhaps more likely that there are different
instantiations of a more general capacity to process and recognize special types of
patterns. It is therefore important to distinguish between a general capacity, and a

122
specific, repeated and (potentially) parallel use of that capacity. I will return to this
important distinction in § 7.5 when I discuss neural redundancy.

7.3 IS THERE A LANGUAGE MODULE?

As we saw in Chapter 4, certain areas of the brain have long been regarded as
quintessentially language areas. For many researchers this assumption and the conviction
that there is far more nature than nurture involved in language acquisition have sat cheek
by jowl. In the last two decades, however, the standard view of how language is organized
and processed in the brain, as well as how it is acquired, has changed dramatically. This is
so for at least two reasons (Kuhl & Damasio 2013). First, neuroimaging evidence in the
form of electroencephalography, magnetoencephalography, positron emission
tomography and (increasingly) functional magnetic resonance imaging has furnished a
wealth of information about how and where language is processed in real time in the
brains of patients carrying out linguistic tasks. The picture which emerges here is very
unlike the one bequeathed by Paul Broca and Carl Wernicke. Second, psycholinguistic
evidence is much richer and more subtle than what was available in previous decades. It
reveals that infants begin learning language from the moment they come into contact with
the sound inventories of their native tongue, indeed, even in utero. It appears that the
early sensitivity of a fetus to features of intonation may later help the infant learn its
mother tongue (Mampe et al. 2009). For instance, the French “papa” has a delayed stress,
and a rising intonation, while the German has an early stress, and a falling intonation.
When an infant begins to form its first sounds, it can build on melodic patterns that are
thus already familiar, and so does not have to start from scratch when learning
phonological and morphological regularities (the investigators suspect the evolutionary
roots of this behaviour to be older than the emergence of spoken language). I shall say a
little more on the acquisition issue in my next section. Here I shall focus on organization,
and review evidence of the extensive reuse of what were traditionally regarded typical
language circuits.

Plausibly, the more distributed a system is in the brain, the more likely it will not
be a specialized system (Anderson 2010; see Chapter 3). It is now known that language is
one of the most distributed systems in the brain, and that “the operation of language to its

123
full extent requires a much more extended network than what [classical models have]
assumed” (Hagoort & Indefrey 2014, p. 359; Anderson 2010, p. 247). As Hagoort and
Indefrey summarize the emerging consensus:

The basic principle of brain organization for higher cognitive functions proposes
that these functions are based on the interaction between numerous neuronal
circuits and brain regions that support various contributing functional
components. These circuits are not necessarily specialized for language but
nevertheless need to be recruited for the sake of successful language processing.
(2014, p. 359)

The evidence motivating this principle in turn corroborates the prediction that more
recently evolved functions should be more distributed than older ones, since it should
overall prove easier to exploit existing circuits than to have to evolve custom-made ones,
with there being “little reason to suppose that the useful elements will happen to reside in
neighboring brain regions….[A] more localist account of the evolution of the brain
would…expect the continual development of new, largely dedicated neural circuits” for
new functions (Anderson 2010, p. 246). Anderson’s review of some 1500 subtraction-
based fMRI experiments suggests that language could well be the paradigm of distributed
processing, supported by more distributed activations than visual perception and
attention (Anderson 2007a) and indeed any other domain that was tested, including
reasoning, memory, emotion, mental imagery and action (Anderson 2008).

Broca’s area holds a special place in the tradition of modular theorizing about
language. While it cannot be doubted that the area plays a crucial role in language
processing, as it happens it is also implicated in various action- and imagery-related tasks
such as those involving the preparation of movement (Thoenissen et al. 2002), the
sequencing of actions (Nishitani et al. 2005), the recognition of actions (Decety et al.
1997; Nishitani et al. 2005), imagery of motion (Binkofski et al. 2000), and the imitation
of actions (Nishitani et al. 2005). It is also known to be involved in certain memory tasks
(Kaan & Stowe 2002) as well as in music perception (Maess et al. 2001). Kaan and Swaab
(2002) set out to identify whether syntactic processing is localized in the brain, and found
that while Broca’s area is recruited during syntactic processing tasks, it joins a larger brain
network that includes the anterior, middle and superior areas of the temporal lobes, none
of which in turn appear to be syntax-specific.

124
 In the auditory domain, phoneme discrimination has long impressed perceptual
psychologists. It involves “categorical perception,” i.e. “the segmenting of a signal that
varies continuously along a number of physical dimensions…into discrete categories, so
that signals within the category are counted as the same, even though acoustically, they
may differ from one another more than do two signals in different categories” (Cowie
2008). Fiona Cowie, no fan of linguistic nativism, accepts that there is a “quite
substantial…inborn contribution to phonological learning.” But, as she goes on to
discuss:

…is this inborn contribution to phonological learning language specific[?]….[T]o

this question, the answer appears to be “No.” First, the “chunking” of
continuously varying stimuli into discrete categories is a feature not just of speech
perception, but of human perception generally. For instance, it has been
demonstrated in the perception of non-linguistic sounds, like musical pitch, key
and melody, and meaningless chirps and bleats…It has also been demonstrated in
the processing of visual stimuli like faces....Secondly, it is known that other
animals too perceive categorically. For instance, crickets segment conspecific
songs in terms of frequency…swamp sparrows “chunk” notes of differing
durations….[O]ther species respond categorically to human speech!
Chinchillas…and cotton-top tamarins…make similar phonological distinctions to
those made by human infants. (Cowie 2008, § 3.3.4)

A very recent experiment found that early exposure to multiple languages heightens
acoustic sensitivity generally (Liu & Kager 2016). In particular, bilingual children appear
more sensitive to subtle variations in musical pitch than their monolingual counterparts.

There is something especially piquant in discovering that classic sensory and

motor areas play a key role in higher thought. In Chapter 2 I reviewed evidence of the
role of vision in semantics. Damasio and Martin demonstrated over two decades ago that
visual areas are active during noun processing tasks (e.g. naming colours, animals, etc.)
(Damasio & Tranel 1993; Damasio et al. 1996; Martin et al. 1995; 1996; 2000). We saw
that word generation in sighted subjects depends at least in part on the bilateral occipital
cortices, regions that have always been thought to be the most specialized in the brain
(Pascual-Leone et al. 2005, p. 394). Beyond the association with phylogenetically older
sensory and perceptual functions, language also seems to have been originally bound up
with the motor system, for motor circuits still appear to be crucial to language perception
and comprehension on many levels of processing (as indeed the functional profile of

125
Broca’s area would tend to suggest). Pulvermüller and Fadiga (2010) report that at the
level of speech perception and processing, changes in the motor and premotor cortex lead
to deficits in phoneme discrimination (2010, pp. 353-355). There is also evidence that the
acoustic properties of phonemes have been shaped to some extent by postural aspects of
the motor system (Graziano et al. 2002; MacNeilage 1998). At the level of semantic
comprehension, magnetic stimulation of the motor system influences the recognition of
semantic word categories (Pulvermüller & Fadiga 2010, pp. 355-357). Pulvermüller
(2005) earlier reported evidence that hearing the words “lick,” “pick” and “kick,” in that
order, activates successively more of the primary motor cortex, suggesting both that the
motor regions involved are inherent to the comprehension task and that comprehension
may involve some kind of simulation. Glenberg et al. (2008) report similar findings, in
particular how the use-driven plasticity of motor circuits affects abstract and concrete
language processing. A particularly intriguing experiment by Glenberg and Kaschak
(2002) showed that response times in sentence comprehension tasks are longer when the
subject is required to perform actions that oppose the direction of motion implied in the
sentences being heard. On its own this is compatible with motor circuits being recruited
via simulation, but actually the result held up even when the sentences related abstract
actions, such as “he sold his house to you,” rather than simply when they involved
obviously evocative sentences such as “put a grape in your mouth” or “you gave the paper
to him.” This argues for more than just simulation, and indicates either a metaphorical
mapping between domains or perhaps some other noncontingent contribution of the
motor cortex in semantic comprehension. Incidentally, it has been demonstrated that
reading comprehension improves when children are allowed to manipulate physical
objects (Glenberg et al. 2007). Finally, syntactic processing seems to depend in important
ways upon the perisylvian cortex, which is involved in the processing of hierarchically
structured action sequences (e.g. lifting a cup, turning it this way, etc., as guided by the
overall aim of quenching thirst) (Pulvermüller & Fadiga 2010, pp. 357-358). And it is
known that both word- and object-combining have overlapping neural implementations
(Greenfield 1991). (I review more evidence of the motor-syntax connection in my
discussion of sequence learning, below.) Taken together, these results strongly suggest
that the motor system enters crucially into the perception and comprehension of language
at various levels of processing, including phonological, semantic and syntactic levels.

126
 This brings us back to Broca’s area. I have already reviewed evidence attesting to
its functional complexity and its importance in action sequencing. A natural response of
those committed to the specificity of language circuits would be to concede all of this
reuse, and say simply that what we are witnessing is the reuse of linguistic circuits for
other, nonlinguistic functions. Given that the motor system is very much older than the
language faculty, the proposal need not be seriously entertained. However, a more subtle
variant of the idea lies behind the contention that Merge may be the source of
productivity and generativity in nonlinguistic domains. As Brattico and Liikkanen pose
the issue:

To how many cognitive domains can this combinatorial operation be applied? In

principle, there seems to be no limit, provided that the appropriate interface
mechanisms are in place. This architecture of language makes it easy to imagine a
recursive symbol processor which can create productive behavior in several
cognitive domains depending on which type of symbols it applies to and which
type of interfaces it is required to handle. (2009, p. 262)

Following Chomsky, they opine that it might have been the application of Merge to
concepts which yielded the “explosive growth of the capacities of thought…leading to the
liberty of the imagination to transpose and change its ideas,” which, as suggested by
Hume, could generate such imaginary objects as “winged horses, fiery dragons, and
monstrous giants.” When Merge is emptied of all content, the result is the system of
natural numbers. 5 And so on. The research above, highlighting the indispensable
contribution of primitive sensory-motor areas for syntactic and semantic processing,
suggests that the argument is skewed, for it tends to imply that Merge, recursion,
metarepresentation or whatever generative engine happens to be invoked to account for
linguistic productivity—with Broca’s area providing its most likely neurological basis (see
e.g. Brattico & Liikkanen 2009, p. 273)—is some sort of ELU, i.e. an integrated,
dedicated, self-contained computational mechanism, perhaps dissociable from core motor
operations (see e.g. Berwick & Chomsky 2016, pp. 75-77). If far more evolutionarily
primitive mechanisms are behind crucial aspects of linguistic processing at the highest
levels, this seems very suppositious. It is more plausible (i.e. parsimonious) to assume that
linguistic productivity was assembled from prior sensory-motor materials, with Broca’s

5“It is not hard to show that, if the lexicon is reduced to a single element, then Merge will yield a form of
arithmetic” (Chomsky 2010, p. 53).

127
area providing a rich source of sequence-processing power (see below). That is to say, the
role of Broca’s area in language is evidence that it already performed just the kind of
sensory-motor functions that made it ideal for integration within a larger language
network (Müller & Basho 2004).

This conjecture is rendered more plausible when one reflects further on the deep
connections between syntactic structure and motor sequences. Even in lower organisms,
motion is never haphazard and shambolic; it is always coordinated, structured and
systematic relative to the organism’s aims and the needs of survival. Coordination is
intrinsic to motor function, a basic prerequisite of meaningful action. Basic body acts
form “action chains” of “meaningful goal-directed action sequence[s],” as exemplified in
the drinking-from-a-cup action sequence mentioned earlier (Pulvermüller and Fadiga
2010, p. 357). A centre-embedded sentence (The man {whom the dog chased away} ran
away) parallels the nested structure of a typical jazz piece (theme {solo} modified theme)
and the action chain formed when entering a dark room at night (open the door {switch
on the light} close the door); in each of these cases, “a superordinate sequence surrounds
a nested action or sequence” (Pulvermüller and Fadiga 2010, p. 357). Indeed the patterns
of coordination and subordination within many complex/cumulative sentences are often
deliberately designed to evoke the actions they describe, a device familiar to writers and
on display in the best literature (Landon 2013). It should not come as a surprise, then,
that syntax recruits the same areas of the brain that are essential for the planning and
coordination of movement. Christiansen and Chater (2016) go a little further, placing
sequence learning at centre-stage of their account of linguistic productivity. They think
complex sequence learning amply explains our ability to process recursive structures, and
(consistent with my theme) that recursivity “relies on evolutionarily older abilities for
dealing with temporally presented sequences of input” (2016, p. 204). There is a wealth
of comparative and genetic evidence—quite apart from the neural evidence I have dwelt
on up to this point—that can also be marshalled in support of the idea that language
makes heavy demands on our complex sequence learning abilities. What is currently
known of the FOXP2 gene is consistent with a human adaptation for sequential
processing (Fisher & Scharff 2009). It is well known that mutations of the gene produce
severe speech and orofacial impairments (Lai et al. 2001; MacDermot et al. 2005).
Moreover when the homologous gene was inserted into mice, the mice displayed superior

128
learning abilities for action sequences (Schreiweis et al. 2014). Specific language
impairment (SLI), for its part, seems to be the result of a clear sequence processing deficit
(Hsu et al. 2014). Further neural evidence of a shared basis for language and general
sequence learning is also available (see the review in Christiansen & Chater 2016, pp. 206-
207). For example, syntactic and sequencing abilities do not appear to dissociate: when
one gets knocked out, chances are the other does too.

Notice by the way that (without prejudging the issue) this account is perfectly
compatible with the idea that aspects of arithmetic, conceptual thought, musical syntax,
and so on, could be exaptations of prior sequence learning capabilities, whether via
language or some other (perhaps more direct) phylogenetic route. Certainly recursive and
metarepresentational capacities seem to crop up elsewhere in cognition, well outside the
domains of language and thought, for example mental time travel, theory of
mind/sociality, culture and morality (Suddendorf 2013).

Thus far I have been largely concerned with the neuroimaging and biobehavioural
evidence against linguistic modularity. For the remainder of this section I shall very
briefly mention a few arguments founded on other considerations, namely, those arising
from evolutionary theory, computational modeling and work on Universal Grammar. In
the upcoming section I shall address the matter of innateness. The final section
introduces my Redundancy Model to account for the rare but still important evidence of
cognitive dissociations, as well as other phenomena not easily explicable without some
such account.

It is widely accepted that of all human phenotypes language is one of relatively

recent origin, certainly far more recently evolved than basic sensory-motor, memory and
conceptual systems. Even if one adopts the view that language and the physiological
mechanisms required to support complex vocalizations evolved together (i.e. that
language and speech co-evolved), by any account language is a phylogenetically recent
phenomenon—de Boer (2016) thinks it is as old as the adaptations for complex
vocalizations and places its emergence at around 400,000 years ago. This fact at once
suggests that specific cognitive adaptations for language are unlikely, essentially for the
reasons already given: it is generally easier for evolution to reuse and exapt existing

129
resources than to have to evolve them anew from scratch (Anderson 2010). But other
reasons support this conclusion as well. For adaptations to arise, evolution requires a
stable environment (Sterelny 2006; Christiansen & Chater 2016). An adaptation for
language would require a linguistically stable environment, but language and cultural
environments generally are anything but stable, with both words and structural features
of languages subject to swift changes, and cultures subject to significant shifts of
convention, often even intragenerationally (Dunn et al. 2011; Greenhill et al. 2010;
Sterelny 2012). In fact when it comes to cultural environments, plasticity is typically
favoured over robustness—changes that allow the organism to cope with unpredictable
variations in the local environment are favoured over specific adaptations narrowly
tailored to that environment, unless of course the culture does provide a stable target over
which selection can operate (see § 7.4, below).

Changing tack somewhat, advances in computational neuroscience have

uncovered a core set of standard, “canonical” neural computations. These computations
are “combined and repeated across brain regions and modalities to apply similar
operations to different problems” (Carandini 2015, p. 179). One example of a canonical
computation, particularly in sensory systems, is “filtering.” This is a basic connectionist
operation in which neurons perform a weighted sum on sensory inputs. The weights are
called “receptive fields,” and the process is performed across the visual, auditory,
somatosensory and possibly motor systems—systems most of which we have seen are
important and even crucial to language processing. Another canonical computation is
“divisive normalization.” This involves dividing neuronal responses by a common factor,
namely the summed activity of a specific collection of neurons. The process is considered
important to operations as varied as “the representation of odours, the deployment of
visual attention, the encoding of value, and the integration of multisensory information”
(Carandini 2015, p. 180). Other examples would include predictive coding, which has
certainly received its fair share of attention in recent years (Clark 2013), as well as
“exponentiation, recurrent amplification, associative learning rules, cognitive maps,
coincidence detection, top-down gain changes, population vectors, and constrained
trajectories in dynamical systems” (Carandini 2015, p. 180). What all this shows is that, at
levels of explanation not too far down—we are still at the “algorithmic” level here, not
quite yet at the circuit or cell level—there are fundamental computations intrinsic to the

130
functioning of the brain which cut across various modalities, very likely including
language. Cognitive operations thus look set to share many of their underlying
computations with other domains, even with domains whose physical resources they do
not actually share.

While we are on the topic, it might be just as well to mention Spaun again (the
brain simulation we met in Chapter 3). Spaun makes a different point, one which I have
been at pains to make in this section, this Chapter, and indeed throughout the whole
thesis. Spaun has been successful in showing that a computer can employ fully domain-
general learning principles, for Spaun reuses the same circuits to accomplish very
different functions (cf. Pinker 1994). As I explained in Chapter 3, most machines are
good at doing just one thing (playing chess, solving equations, etc.). Spaun is unique both
as to the variety of the tasks it can perform and in its ability to learn new tasks using the
same set of circuits. It is the first major step in answering an important challenge leveled
by evolutionary psychologists and other proponents of traditional forms of modularity
who for many years said that such a machine could not be designed (virtually on a priori
grounds!). Well, Spaun is a machine that functions entirely by domain-general principles.
(QED.)

Finally, a word on Universal Grammar. One of the enduring ideas that has fallen
out of the generative tradition, and to which a majority of both adherents and detractors
alike have probably subscribed at one time or another, is the idea that all human
languages share the same underlying grammatical base despite surface differences. In
recent years, however, Universal Grammar has been discomfited by evidence of the
bewildering variety of languages divergent in so many points of typology, structural and
otherwise, as to render the theory of a universal syntax highly dubious. A point Evans and
Levinson made early on in their 2009 review of the subject bears repeating here, for it is
perhaps the single most stunning fact about the whole affair (the very idea, the project,
the prospects, etc.):

Somewhere between 5,000 and 8,000 distinct languages are spoken today….Less
than 10% of these languages have decent descriptions….If we project back
through time, there have probably been at least half a million human
languages…so what we now have is a non-random sample of less than 2% of the
full range of human linguistic diversity. It would be nice to at least be in the
131
 position to exploit that sample, but in fact, as mentioned, we have good
information for only 10% of that. (Evans & Levinson 2009, p. 432)

Thus “nearly all generalizations about what is possible in human languages” are based on
a pitiful 10% of less than 2% of all the languages that have ever existed on Earth! This is
somewhere between sobering and depressing, especially given that “language death
continues apace.” Universal Grammar does not support the case for an ELU.

7.4 IS LANGUAGE INNATE?

Let me begin with a couple of straightforward observations. Infants begin their lives with
a remarkable ability to detect and respond to subtle acoustic distinctions that vary
considerably across the world’s languages. Within a short time—indeed before 10 months
of age (Kuhl & Damasio 2013)—and pursuant to a powerful and in some respects still
mysterious learning process, they come to recognize statistical properties in the acoustic
stream, form phonetic categories, distil words and possibly inflectional items, and
assimilate the basic phrase structure of their mother tongue. By one year of age they
appear able to comprehend simple imperatives like “Show me your nose” (Glickstein
2014). And even though the linguistic environment is not as impoverished as was once
believed (Clark 2009, p. 368; Pullum & Scholz 2002; Scholz & Pullum 2002), it is still
striking that most if not all of this gets underway without explicit instruction or drilling.
Overall, adopting the terminology and concepts introduced in Chapter 6, let us admit that
language acquisition involves a certain degree of developmental robustness—not quite
like that of the visual system or the growth of wings on birds (or the growth of limbs, or
the onset of puberty, or any of the other rhetorical claims made in the past), but
something to reckon with nonetheless. The beauty of developmental robustness is that it
admits of degrees, so that to confess that a system is characterized by robustness does not
commit one to implausible claims. Actually developmental robustness is somewhat
reminiscent of the notion of being acquired under a “poverty of stimulus,” because to be
so acquired just is to develop independently of the presence of some specific
environmental stimulus or stimuli, and hence denotes a sort of invariance with respect to
experience (Griffiths & Machery 2008, pp. 406-407). Drawing this link is acceptable so
long as the locution is employed with care, and in the understanding that acquisition
under poverty of stimulus is a relative phenomenon, not an absolute one (see Chapter 6).

132
 The question now is whether this developmental profile reopens the debate over
the existence of an ELU. Does the fact that language seems to be acquired so early in life,
with fair uniformity over substantial variations of intelligence and experience, without
specific training, and so on, call for the postulation of a domain-specific language module?
I think this is a reasonable move, but an unnecessary one nonetheless. It is worth
remembering that however robust language acquisition might be, even in the most ideal
conditions it can take a long time to complete (up to ten years or more). There is anyway a
more natural and parsimonious explanation for why language acquisition proceeds at the
ontogenetic pace it does, and why it often seems that children attain mastery of their
native language almost effortlessly. The explanation lies in the mutual accommodation (or
fit) between the processing dispositions of the brain regions used in language and
language itself. There is evidence both that language was culturally shaped (as a “cultural
tool”) to be learnable and easy to process through cultural evolution (Everett 2012;
Christiansen & Chater 2016; Laland 2016) and that selective pressures in the course of
biological evolution may have equipped the brain with the sorts of processing dispositions
and biases that made language easier to learn and process (Dor & Jablonka 2010; Sterelny
2006; Christiansen & Chater 2016; see also Laland et al. 2015). Since cultural evolution is
much the most important side of this story, I shall have a little more to say about that
than about biological evolution. But before I go any further, let me frame the main point
of this section in terms of Stanislas Dehaene’s (2005) “neuronal recycling” hypothesis,
which we met briefly in Chapter 3.

Recall that towards the end of Chapter 6 I observed that while there is a relative
sense in which modules and other functionally significant brain regions can be considered
innate, the same cannot be said for the high-level cognitive functions composed of them.
Cast in terms of reuse, low-level cognitive workings may be innate, but it does not follow
from this that high-level cognitive uses are innate. Most complex cognitive functions are
learned throughout the course of a person’s life—whether it be riding a bicycle, tying
one’s shoes, or reading, these skills do not spontaneously unfurl as a result of intrinsically
determined developmental processes, so it makes sense to withhold the designation
“innate” or “robust” from the C-networks which implement them. Why then is language
acquisition different from reading, performing long division, or learning physics? What is
it about conversation that entitles us to regard it (and its C-network) as in some sense

133
sharing in or inheriting the robustness of its components (workings/M-networks)? This
is where Dehaene’s notion of a “neuronal niche” is useful. Cultural acquisitions must
make their home among a particular ensemble of cortical regions (a C-network) and this
process is akin to the process of organisms creating their own ecological niches among the
habitats in which they find themselves. Just as organisms must make the best use of the
resources at their disposal, so cognitive organisms (i.e. cultural acquisitions) are
constrained by the processing dispositions of the brain regions required for the tasks at
hand. We have already seen that brain regions do have robust processing capabilities and
clear input preferences. Dehaene’s idea is that the more the acquired practice matches the
processing dispositions of the brain regions recruited for the task, the easier and less
disruptive the learning process because the neural composite does not require a radical
departure from existing cortical biases. On the other hand, the greater the distance
between the acquired practice and the processing dispositions of the brain regions it will
draw upon, the more difficult and protracted the learning process will be, potentially
disrupting the regions’ established operations and whatever functional composites they
already subserve.

It is not hard to see how this account would dovetail nicely with a cultural
evolutionary account revealing the ways in which language has been shaped over many
hundreds of generations to be learnable and easy to process. If human languages have in
fact been so worked upon as to make them easy to learn and use, the neuronal niche
which languages must nuzzle into already ideally conforms to the sorts of processing
demands that languages impose on language users. It is just as well, then, that there is just
such a cultural story to tell! Brighton et al. (2005) call it “cultural selection for
learnability.” If the rudiments of syntax, phonology, morphology and so on are to survive
from one generation to the next, they must earn their keep. If they are too cumbersome or
exotic to be readily learned, taken up and transmitted to the next generation, they will be
discarded for simpler and more streamlined or efficient devices. There is mathematical
modeling to suggest that compositionality could have evolved in this fashion, for instance
(Smith & Kirby 2008; Kirby et al. 2007).

It is hard to deny that human languages are cultural products (Everett 2012). And
if so, it makes perfect sense that they will reflect the cognitive and neural dispositions of

134
the agents which created them in their own image. As Christiansen & Chater (2016, pp.
43-44) explain:

In other cultural domains, this is a familiar observation. Musical patterns appear

to be rooted, in part at least, in the machinery of the human auditory and motor
systems…art is partially shaped by the properties of human visual
perception…tools, such as scissors or spades, are built around the constraints of
the human body; aspects of religious beliefs may connect, among other things,
with the human propensity for folk-psychological explanation.

They identify and elaborate upon four groups of nonlinguistic constraints that they
conjecture would have guided the cultural evolution of language. (Much of this can be
read as a natural extension of the ideas in § 7.3 concerning the reuse of language circuits.)
They divide the constraints here between those arising from thought, perceptuo-motor
factors, memory and pragmatics. For example, on the assumption that thought is “prior
to, and independent of, linguistic communication,” key properties of language such as
compositionality, predicate-argument structure, quantification, aspect and modality can
be “proposed to arise from the structure of the thoughts language is required to express”
(2016, p. 51). Cognitive linguists have made the dependence of language on thought a
critical feature of their perspective, arguing that our basic conceptual repertoire,
including space and time, can be seen to have left their mark on the structure and
categories of the world’s languages (Croft & Cruise 2004; Evans & Green 2006).
Perceptuo-motor constraints have also left their mark, most obviously in “the seriality of
vocal output” which “forces a sequential construction of messages” (2016, p. 52).
Christiansen and Chater speculate that

The noisiness and variability…of vocal…signals may, moreover, force a “digital”

communication system with a small number of basic messages: e.g., one that uses
discrete units (phonetic features, phonemes, or syllables). The basic phonetic
inventory is transparently related to deployment of the vocal apparatus, and it is
also possible that it is tuned, to some degree, to respect “natural” perceptual
boundaries. (2016, p. 52)

The extent of the connections here can be taken quite far. MacNeilage (1998), for
example, has offered the intriguing hypothesis that syllabic structure might have been
partly determined by the jaw movements involved in mastication! While not immediately
obvious, on reflection it seems likely that many complex aspects of phonology and

135
morphology will be traced to similarly prosaic origins. Memory constraints are hardly less
significant, seen, for instance, in the tendency to resolve linguistic dependencies (e.g.
between arguments and their verbs) as early as possible, “a tendency that might not be
syntax-specific, but instead an instance of a general cognitive tendency to attempt to
resolve ambiguities rapidly whether for linguistic…or perceptual input (Christiansen &
Chater 2016, p. 53). Finally, pragmatic constraints must have wielded a hefty influence
on many aspects of language design—Levinson (2000) showed that discourse and
anaphora appear to be related, so it is plausible that aspects of binding theory could be
accounted for in terms of pragmatics. In all these ways and without doubt very many
more (including ways yet to be explored—a monumental undertaking really) language has
been “shaped by the brain,” naturally and parsimoniously explaining the child’s relative
ease of acquisition and the intimate relationship between the child’s innate endowment
and the structure of language.

Before concluding this section, I should indicate something of the process of

mutual fit and accommodation as it occurs in the other direction. While language has
been predominantly shaped by the brain, to be sure, in certain limited respects it is at
least likely that the brain has been shaped through selection pressures for language. In the
previous section I mentioned that in order for adaptations to arise, evolution requires a
stable environment, and that adaptations for language specifically would require a
linguistically stable environment. I also said that linguistic and other cultural
environments are in the nature of things quite unstable, and that given these
contingencies, when it comes to cultural environments, plasticity is typically favoured
over robustness. This is just to say that unstable environments are conducive to the sorts
of nervous systems which exploit the same resources for alternative ends, so that the
cognitive mechanisms which are selected for in such circumstances will typically be
flexible enough to be put to alternative uses (Avital & Jablonka 2000; Dor & Jablonka
2010). Now is as good a time as any to reference the well-known phenomenon of “niche
construction,” part of the “Extended Synthesis” in evolutionary biology (see Laland et al.
2015 and Laland et al. 2011 for reviews). Niche construction is a specific instance of the
broader process of gene-culture coevolution (Boyd et al. 2011; Richerson & Boyd 2005).
The essential difference is that the process is cumulative. Organisms are always altering
their environments to better suit their needs, whether by creating nests, burrows, dams

136
and so on. In the case of humans, these environmental modifications extend to the social
and cultural worlds that encompass language. The changes wrought in these ways
inevitably modify the selection pressures acting on organisms and so facilitate adaptation
to the new environments they have created, which organisms will inevitably alter further
still, which leads to new selection pressures, and so on and on in a virtuous cycle of
organism-directed environmental and cultural modification and adaption that results in
organisms being increasingly better adapted to the material, social and cultural worlds of
their own making. It is highly likely that cognitive mechanisms evolved for language in
this manner (not ELU’s, however: see following), particularly to the extent that we can
identify universal, stable features across linguistic environments (such as a constrained
range of phonemic units arranged combinatorially and with duality of patterning). Laland
(2016, p. 5) conjectures that “[i]mportant elements of infant-directed speech, such as
infants’ sensitivity to its linguistic features, or adults’ tendency to engage in behaviour
that elicits rewarding responses from infants (e.g. smiles), have been favoured through a
biological evolutionary process.” Adding to the list of adaptations which would have been
crucial in the evolution of a language faculty we could cite the ability to represent
symbolically (Deacon 1997), the ability to reason about other minds (Malle 2002), the
ability to engage in pragmatics (Levinson 2000), increased working memory (Gruber
2002), an increased domain-general capacity for learning words (Bloom 2000) and
modifications to the human vocal tract (descended larynx, etc.) (de Boer 2016). It is vital
to stress that in respect of none of these adaptations can we say that we are we dealing
with an ELU—language may have provided the occasion for selection, but there is no
evidence that these mechanisms are used exclusively for language, and indeed
overwhelming evidence that the brain simply “doesn’t work that way”: virtually no
cortical structure, not even the visual cortex(!), is so insensitive to experience that it
resists all cooption during development. Rather, the evidence points to a brain that
integrates all sorts of brain regions within the neural ecology for the management of
organism-environment interactions, even where these regions might by nature be
disposed to processing particular sorts of inputs over others. This makes good
evolutionary sense, being overall “a more efficient use of metabolically expensive brain
matter” (Anderson 2014, p. 46). Even the structure of the vocal apparatus has uses
outside the language faculty (in music and meditation for example).

137
 One last thing: cognitive adaptations of relevance to a specific domain like
language may require no more than a simple change to synaptic connection patterns, for
instance a genetic event that entrenches a pattern of connections between a set of
preexisting domain-general modules (Ramus & Fisher 2009, p. 865). This is in fact just
what the theory of reuse entails, at least for many cases involving the emergence of novel
traits—to the extent that the theory holds that it will often be easier to mix and match
existing elements than to have to evolve them afresh each time a new evolutionary
challenge arises, the theory implies that specific combinations of neural elements (which
have perhaps proved their value developmentally) will be selected for. How else can a
specific arrangement of preexisting domain-general modules be entrenched other than
through a robust synaptogenetic process of some description or another (see § 7.5, below,
on “search”)? Thus it could be that some parts of the language C-network, perhaps even
large parts, are already wired up and ready to go, even though the modules within the
network are entirely domain-general. Preformed connections would surely result in a
smooth period of language learning, even given “relatively slight exposure and without
specific training” (Chomsky 1975, p. 4).

7.5 ACCOUNTING FOR LINGUISTIC MODULARIZATION

Throughout this chapter I have been investigating a very particular question, and an
important one: does language rely on specialized cognitive and neural machinery, or does
it rely on the same machinery that allows us to get by in other domains of human
endeavour? The question is bound up with many other questions of no less importance,
questions concerning the uniqueness of the human mind, the course of biological
evolution and the power of human culture. What is perhaps a little unusual about this
question, however—unusual for a question whose answer concerns both those working in
the sciences and the humanities—is that it can be phrased as a polar interrogative, i.e. as a
question which admits of a yes or no response. And indeed the question has divided
psychologists, linguists and the cognitive science community generally for many decades
now, more or less into two camps. In this concluding section I would like to sketch the
beginnings of an answer to this question in a way that does not pretend it can receive a
simple yes or no. Let me stress again that neural reuse is undeniable, that the evidence for
it is simply overwhelming, and that it has left no domain of psychology untouched. There

138
seems to be nothing so specialized in the cortex that it cannot be repurposed to meet new
challenges. In that regard, to be sure, what I am proposing in this section is
unapologetically on the side of those who maintain that language is not special—that
there is no specialized “language organ” or ELU. And yet I would like to carefully
distinguish this claim from the claim that there are no areas of the brain that subserve
exclusively linguistic functions. The neuropsychological literature offers striking
examples of what appear to be fairly clean dissociations between linguistic and
nonlinguistic capacities, i.e. cases in which language processing capacities appear to be
disrupted without impeding other cognitive abilities, and cases in which the reverse
situation holds (Fedorenko et al. 2011; Hickok & Poeppel 2000; Poeppel 2001; Varley et
al. 2005; Luria et al. 1965; Peretz & Coltheart 2003; Apperly et al. 2006). An example
would be where the ability to hear words is disrupted, but the ability to recognize non-
word sounds is spared (Hickok & Poeppel 2000; Poeppel 2001). Discussing such cases,
Pinker and Jackendoff (2005, p. 207) add that “[c]ases of amusia and auditory agnosia, in
which patients can understand speech yet fail to appreciate music or recognize
environmental sounds…show that speech and non-speech perception in fact doubly
dissociate.” Although as we saw in Chapter 4 dissociations are compatible with reuse—
indeed there is work suggesting that focal lesions can produce specific cognitive
impairments within a range of nonclassical architectures (Plaut 1995)—and it is equally
true that often the dissociations reported are noisy (Cowie 2008, § 3.6.3), still their very
ubiquity needs to be taken seriously and accounted for in a more systematic fashion than
many defenders of reuse have been willing to do (see e.g. Anderson 2010, p. 248; 2014,
pp. 46-48). After all a major source of support for theories of reuse comes from the
neuroimaging literature, which, as I have pointed out several times already, is somewhat
ambiguous taken by itself. As Fedorenko et al. (2011, p. 16428) explain:

standard functional MRI group analysis methods can be deceptive: two different
mental functions that activate neighbouring but non-overlapping cortical regions
in every subject individually can produce overlapping activations in a group
analysis, because the precise locations of these regions vary across subjects,
smearing the group activations. Definitively addressing the question of neural
overlap between linguistic and nonlinguistic functions requires examining overlap
within individual subjects, a data analysis strategy that has almost never been
applied in neuroimaging investigations of high-level linguistic processing.

139
When Fedorenko and her colleagues applied this strategy themselves, they found that
“most of the key cortical regions engaged in high-level linguistic processing are not
engaged by mental arithmetic, general working memory, cognitive control or musical
processing,” and they think that this indicates “a high degree of functional specificity in
the brain regions that support language” (2011, p. 16431). While I do not believe that
claims of this strength have the least warrant—as I shall explain, functional specificity
cannot be established merely by demonstrating that a region is selectively engaged by a
task—these results do at least substantiate the dissociation literature in an interesting way
and make it more difficult for those who would prefer to dismiss the dissociations with a
ready-made list of alternative explanations. Similar results were found by Fedorenko et
al. (2012).

I think neural redundancy is the best explanation for what we see in cases like
these, and that redundancy is in fact a central feature of cortical design. As I briefly
mentioned in Chapter 6, the brain incorporates a large measure of redundancy of function
(Laurence & Margolis 2015, p. 126; Jungé & Dennett 2010, p. 278; Barrett & Kurzban
2006, pp. 638-639; Anderson 2010, p. 296). Modules (M-networks) and similar structures
in the brain fall in an iterative, repetitive and almost lattice-like arrangement in the
cortex. Neighbouring modules have similar response properties: laminar and columnar
changes are for the most part smooth—not abrupt—as one moves across the cortex, and
adjacent modules do not differ markedly from one another in their basic structure and
computations (if they even differ at all when taken in such proximity). Regional solitarity
is therefore not likely to be a characteristic of the brain (Anderson 2014, p. 141).6 We do
not, in all likelihood, have just one module for X, and one module for Y, but in effect
several copies of the module for X, and several copies of the module for Y, all densely
stuffed into the same cortical zones. As Buxhoeveden and Casanova (2002, p. 943) explain
of neurons generally:

In the cortex, more cells do the job that fewer do in other regions….As brain
evolution paralleled the increase in cell number, a reduction occurred in the
sovereignty of individual neurones; fewer of them occupy critical positions. As a
consequence, plasticity and redundancy have increased. In nervous systems

6The term “solitarity” is Anderson’s, but while he concedes that solitarity will be “relatively rare,” he does
not appear to believe that anything particularly significant follows from this. See also Anderson (2010, p.
296).

140
 containing only a few hundred thousand neurones, each cell plays a more essential
role in the function of the organism than systems containing billions of neurones.

The same principle very likely holds for functionally distinct groupings of neurons (i.e.
modules), as Jungé and Dennett conjecture:

It is possible that specialized brain areas contain a large amount of

structural/computational redundancy (i.e., many neurons or collections of
neurons that can potentially perform the same class of functions). Rather than a
single neuron or small neural tract playing roles in many high-level processes, it is
possible that distinct subsets of neurons within a specialized area have similar
competencies, and hence are redundant, but as a result are available to be assigned
individually to specific uses….In a coarse enough grain, this neural model would
look exactly like multi-use (or reuse). (2010, p. 278)

This is plausibly why capacities which are functionally very closely related, but which for
whatever reason are forced to recruit different neural circuits, will often be localized in
broadly the same regions of the brain. For instance, first and second languages acquired
early in ontogeny settle down in nearly the same region of Broca’s area; and even when
the second language is acquired in adulthood the second language is represented nearby
within Broca’s area (while artificial languages are not) (Kandel & Hudspeth 2013). The
neural coactivation graphs of such C-networks must look very similar. Indeed these
results suggest—and a Redundancy Model would predict—that two very similar tasks
which for whatever reason are forced to recruit different neural circuits should exhibit
similar patterns of activation.

The significance of this simple but surprisingly neglected feature of cortical

design cannot be overstated. For reasons I shall explain, I think it should rank alongside
reuse as an organizing principle of the brain. What it means for reuse is quite interesting.
Although there is abundant evidence of the reuse of the same neural tokens to accomplish
different tasks (see Chapters 3 and 5), redundancy means we must accept that at least
some of the time what we will really be witnessing is reuse of the same types to
accomplish these tasks.7 To my mind this does not in any way diminish the standing of
reuse. To the extent that a particular composite reuses types, and is pro tanto
dissociable—residing in segregated brain tissue that is not active outside the function

7For a developmental twist on the type/token distinction invoked in the context of modular theorizing
about the mind, see Barrett (2006).

141
concerned—it is true that to that extent its constituents will appear to be domain-specific.
But in this case looks will be deceiving. The classical interpretation of domain specificity
assumes solitarity—that a module is unique in what it contributes (that a module for X
does something which no other module can do as well). Dedication is not the same as
domain specificity, and redundancy explains why. A composite will be dedicated without
being domain-specific if its functional resources are accessible to other domains through
the deployment (reuse) of neural surrogates (i.e. redundant or “proxy” tokens). In this
case its constituents will be multi-potential but single-use (Jungé & Dennett 2010, p.
278), and the domain specificity on display somewhat cosmetic. For instance, a set of
modules that are structurally and computationally similar may be equally suited for face
recognition tasks, abstract-object recognition tasks, the recognition of moving objects,
and so on. One of these modules could be reserved for faces, another for abstract objects,
another for moving objects, and so on—what is noteworthy is that while the functional
activation may be indistinguishable in each case, and the same type of resource will be
employed on each occasion, a different token module will be at work at any one time. To
quote Jungé and Dennett again:

In an adult brain, a given neuron [or set of neurons] would be aligned with only a
single high-level function, whereas each area of neurons would be aligned with
very many different functions. (2010, p. 278)

To borrow a political slogan, such modules (and composites) will be “separate but equal.”
The evidence of reuse is virtually all one way when it comes to the pervasiveness of
functional inheritance across cognitive domains. It may be that this inheritance owes to
reuse of the same tokens (literal reuse) or to reuse of the same types (reuse by proxy), but
the inheritance itself has been amply attested. This broader notion of reuse still offers a
crucial insight into the operations of cognition, and I dare say represents a large part of
the appeal of the original massive redeployment hypothesis (Anderson 2007c).

It is interesting to note in this respect that although detractors have frequently

pointed out the ambiguity of neuroimaging evidence on account of its allegedly coarse
spatial resolution (see § 3.3.3), suggesting that the same area will be active across separate
tasks even if distinct but adjacent circuits are involved each time, this complaint can have
no bearing whatsoever on reuse by proxy. Fedorenko et al. (2011, p. 16431) take their

142
neuroimaging evidence to support “a high degree of functional specificity in the brain
regions that support language,” but their results do not license this extreme claim. The
regions they found to have been selectively engaged by linguistic tasks were all adjacent to
the regions engaged in nonlinguistic tasks. Elementary considerations suggest that they
have discovered a case of reuse by proxy involving language: the domains tested (mental
arithmetic, general working memory, cognitive control and musical processing) make use
of many of the same computations as high-level linguistic processing, even though they
run them on duplicate hardware. Redundancy makes it is easy to see how fairly sharp
dissociations could arise—knocking out one token module need disrupt only one high-
level operation: other high-level operations that draw on the same type of resources may
well be spared.

The consequences of this distinction between literal reuse and reuse by proxy for
much speculation about the localization and specialization of function are potentially
profound. In cognitive neuropsychology the discovery that a focal lesion selectively
impairs a particular cognitive function is routinely taken as evidence of its functional
specificity (Coltheart 2011; Sternberg 2011). Even cognitive scientists who take a
developmental approach to modularity, i.e. who concede that parts of the mind may be
modular but stress that modularization is a developmental process, concede too much
when they imply, as they frequently do, that modularization results in domain-specific
modules (Karmiloff-Smith 1992; Prinz 2006; Barrett 2006; Cowie 2008; Guida et al.
2016). This is true in some sense, but not in anything like the standard sense, for the
Redundancy Model envisages that developmental modules form a special class of C-
networks, namely those which are “separate but equal.” The appearance of
modularization in development is thus fully compatible with deep domain
interpenetration. In any event the Redundancy Model does not predict that all acquired
skills will be modular. The evidence suggests that while some complex skills reside in at
least partly dissociable circuitry, most complex skills are implemented in more typical C-
networks (i.e. those consisting of literally shared parts).8

Asking why the cortex incorporates a large measure of redundancy of function is a

bit like asking why we have two eyes, two kidneys, ten toes, and so on. The intuitive

8 This seems to be true regardless of whether the complex skills are innate or acquired.

143
response is that by having “spare” organs we can distribute the workload more efficiently
among all of them, and that it is generally a good design feature of any system to have
spare capacity: having back-up cortical tissue makes sense, just like having a spare car tyre
in the boot makes sense. These are good reasons for evolution to have seen to it that our
brains have spare capacity. But in the case of the brain and the cortex most especially, I
think there are other reasons why redundancy would be an important design feature. It
offers a solution to what Jungé and Dennett (2010, p. 278) called the “time-sharing”
problem. It may also offer a solution to what I call the “encapsulation” problem.

The time-sharing problem arises when multiple simultaneous demands are made
on the same cognitive resource. This is probably a regular occurrence, and language in
particular would present a whole host of opportunities for time-sharing. Here are just a
few examples.

• Driving a car and holding a conversation at the same time: if it is true that some of
the selfsame motor operations underlying aspects of speech production and
comprehension are also required for the execution of sequenced or complex motor
functions (as perhaps exemplified by driving a manual vehicle, or operating
complex machinery), how do we manage to pull this off?
• By reflecting the recursive structure of thought, the coding function may redeploy
the recursive operation simultaneously during sentence production. This might
be the case during the formation of an embedded relative clause—the thought and
its encoding may require parallel use of the same sequencing principle. Again,
how do we manage this feat?
• If metarepresentational operations are involved in the internalization of
conventional sound-meaning pairs, and also in the pragmatics and mindreading
that carry on simultaneously during conversation, as argued by Suddendorf
(2013), it could be another instance of time-sharing. The example is contentious,
but it still raises the question: how does our brain manage to do things like this?
• Christiansen and Chater’s (2016) “Chunk and Pass” model of language processing
envisages multilevel and simultaneous chunking procedures. As they put it, “the
challenge of language acquisition is to learn a dazzling sequence of rapid

144
 processing operations” (2016, p. 116). What must the brain be like to allow for
this dazzling display?

Explaining these phenomena is difficult. Indeed when dealing with clear (literal)
instances of reuse, interference paradigms show that processing bottlenecks are
inevitable—true multi-tasking is impossible (see § 4.2.3). Redundancy offers a natural
explanation of how the brain overcomes the time-sharing problem. It explains, in short,
how we are able to “walk and chew gum” at the same time.

Redundancy might also offer a solution to the encapsulation problem. As I

explained in § 4.2.3, functional composites are not likely to be characterized by
informational encapsulation because in sharing their parts with other systems they will
prima facie have access to the information stored and manipulated by those other systems
(Anderson 2010, p. 300). If overlapping brain networks must share information on some
level (Pessoa 2016, p. 23), it would be reasonable to suppose that central and peripheral
systems do not overlap. This is because peripheral systems, which are paradigmatically
fast and automatic, would not be able to process inputs as efficiently if there were a
serious risk of central system override—i.e. of beliefs and other central information
getting in the way of automatic processing. But we know from the neuroimaging
literature that quite often the brain networks implementing central and peripheral
functions do overlap. This is puzzling in light of the degree of cognitive impenetrability
that certain sensory systems still seem to exhibit—limited though it may be. If it is
plausible to suppose that the phenomenon calls for segregated circuitry, redundancy
could feature in a solution to the puzzle, since it naturally explains how the brain can
make parallel use of the same resources. Neuroimaging maps might well display what
appears to be overlapping brain regions between two tasks (one involving central
information, the other involving classically peripheral operations), but the overlap would
not exist—there would be distinct albeit adjacent and nearly identical circuits recruited in
each case. Of course there may be other ways around the encapsulation problem that do
not require segregated circuitry: the nature and extent of the overlap is presumably
important. But clearly redundancy opens up some fascinating explanatory possibilities.

145
 To the extent that acquired skills must overcome both the time-sharing problem
as well as the encapsulation problem—for acquired competencies are often able to run
autonomously of central processes—we might expect that their neural implementations
incorporate redundant tissue. In concluding, let me illustrate this point by offering a gloss
on a particular account of how skills and expertise are acquired during development
elaborated by Guida et al. (2016) and Anderson (2014). The process involved is called
“search.” Search is an exploratory synaptogenetic process, “the active testing of multiple
neuronal combinations until finding the most appropriate one for a specific skill, i.e., the
neural niche of that skill” (Guido et al. 2016, p. 13). The theory holds that in the early
stages of skill acquisition, the brain must search for an appropriate mix of brain areas, and
does so by recruiting relatively widely across the cortex. When expertise has finally
developed, a much narrower and more specific network of brain areas has been settled
upon, such that “[a]s a consequence of their extended practice, experts develop domain-
specific knowledge structures” (Guido et al. 2016, p. 13). The gloss (and my hunch) is
this: first, that repeated practice of a task that requires segregation (to get around time-
sharing and encapsulation issues) will in effect force search into redundant neural
territory (Karmiloff-Smith 1992; Barrett 2006; Barrett & Kurzban 2006); second, that
search will recruit idle or relatively underutilized circuits in preference to busy ones as a
general default strategy. Guido et al. (2016) cite evidence that experts’ brains reuse areas
for which novices’ brains make only limited use: “novices use episodic long-term memory
areas (e.g., the mediotemporal lobe) for performing long-term memory tasks,” but
“experts are able to (re)use these areas also for performing working-memory tasks”
(Guido et al. 2016, p. 14). Guido et al., in agreement with Anderson (2014), seem to have
literal reuse in mind. But the same evidence they cite is consistent with reuse by proxy.
As Barrett and Kurzban (2006, p. 639) suggest, echoing a similar suggestion by
Karmiloff-Smith (1992), a developmental system

could contain a procedure or mechanism that partitioned off certain tasks—

shunting them into a dedicated developmental pathway—under certain
conditions, for example, when the cue structure of repeated instances of the task
clustered tightly together, and when it was encountered repeatedly, as when
highly practiced….Under this scenario, reading could still be recruiting an
evolved system for object recognition, and yet phenotypically there could be
distinct modules for reading and for other types of object recognition.

146
 7.6 SUMMARY

On any reasonable construal of the language faculty, language is not cognitively special
vis-à-vis other cognitive domains. There seems to be no language module, no elementary
linguistic unit, no hardwired language organ. Language was likely assembled from older
sensory-motor and nonlinguistic materials. Neuroimaging, biobehavioural,
computational, typological and evolutionary considerations all point to the same
conclusion. Such linguistic adaptations as there have been have been coopted in many
other domains of cognition. The sort of cultural environment in which language consists
is too unstable to provide the conditions for typical selection scenarios in which robust
phenotypes can emerge, and the brain anyway negotiates energetic constraints by
repurposing existing resources to meet new challenges. Language acquisition frequently
does seem effortless on the child’s part, and exhibits a degree of developmental
robustness. But the ease of acquisition has been exaggerated—the child’s environment is
not as impoverished as was once assumed. In any case this apparent ease can be explained
other than by postulating exotic and impossible-to-evolve circuitry. Language has been
shaped by the brain far more than the brain has been shaped by language. Cultural
evolution is a powerful factor in human history, and is more than sufficient to explain
why languages seem to run so well with the grain of the human mind. It is true that
language dissociates from other cognitive skills, at least in some respects, but the
Redundancy Model puts this sort of modularization in its proper context. The
Redundancy Model predicates functional inheritance across tasks and task categories even
when the tasks are implemented in spatially segregated neural networks. Thus
dissociation evidence alone does not always indicate functional specificity. In particular,
these dissociations provide no evidence that language is cognitively special.

147
 8
_____

Saving faculty psychology:

Debunking the argument from multiple realization

8.1 NEW DIRECTIONS IN FACULTY PSYCHOLOGY

The discovery that traditional psychological faculties are implemented by neural systems
consisting of shared domain-general components does not make traditional faculty
psychology go away. On the contrary, as I argued in § 4.2.3, these high-level cognitive
systems—I decline to call them modules—have an important role to play in our ongoing
quest to understand the mind. The ramifications of reuse will in fact serve to sharpen our
understanding of what makes these systems tick and reveal the extent of functional and
semantic inheritance between traditional tasks and task categories. A mature, twenty-first
century faculty psychology, therefore, has a lot to look forward to—and essentially
nothing to fear. But if the work required to understand behaviour in the light of neural
reuse is going to get done by anyone, it will have to get done (one would think, in the first
instance at least) by twenty-first century faculty psychologists! All those with a wealth of
experience investigating the higher faculties will simply have to get on board if the
endeavour is to have any chance of success. In other words the future of faculty
psychology depends in no small part on the productive collaboration between
neuroscience and psychology.

Unfortunately, there is a potential obstruction in the way of just this sort of

intertheoretic collaboration. Inasmuch as high-level cognitive systems are understood to
be software systems, pitched at the level of algorithmic or computational psychology, the

148
thought arises that these systems can be investigated with a kind of disciplinary immunity
from neuroscience, since (it is supposed) “[n]o amount of knowledge about the hardware
of a computer will tell you anything serious about the nature of the software that
computer runs” (Coltheart 2004, p. 22). The idea proceeds from the assumption that
psychological processes are multiply realized. The multiple realization (“MR”)
hypothesis asserts, at its baldest, that the same psychological state may be realized in
neuroscientifically distinct substrates (Polger 2009). Hilary Putnam’s (1967) ingenious
suggestion that “pain” is likely to be a multiply realized kind (“MR kind”) rather neatly
captures the thought here—while presumably both mammals and mollusks experience
pain, they pretty obviously lack the same neurobiological basis. MR was played against a
popular philosophical theory of mind in the 1960s which attempted to identify mental
states with neural states. Since MR implies a many-to-one mapping from neural states to
mental states, if it is in fact true that mental states are multiply realized it follows that no
clear identity relation can hold between them. Thus it was that many of those who
advanced MR rejected mind-brain identity as a viable philosophical theory. As Bechtel
and Mundale (1999, p. 176) frame the issue, “[o]ne corollary of this rejection of the
identity thesis is the contention that information about the brain is of little or no relevance
to understanding psychological processes.”

Another development of the argument asserts that even if MR does not hold
among existing biological systems, the possibility that cognitive states might be shared by
built artifacts or alien life forms having very different physical structures in itself
establishes the salience of the MR thesis. This version of the argument led some
philosophers of AI to embrace the further metaphysical claim that “mental processes are
the operations themselves, and are not identified with whatever biological or other
substances realize them” (Bechtel & Mundale 1999, p. 176). Following Bechtel and
Mundale’s lead, and the contours of the recent debate surrounding MR, I shall not
address this version of the argument or the metaphysical claim which it inspired here.
The possibility that artifacts could have mental states is just the possibility that the
identity theory is wrong, which is precisely the claim in dispute (Polger 2009, p. 459).
Instead, I shall use this chapter to consider, admittedly quite briefly, the empirical
claim—that the MR hypothesis can be verified having regard to existing organisms—

149
since no doubt it is this claim that has played the lion’s share in encouraging a downbeat
attitude to evidence from neuroscience in some quarters.

Daniel Dennett (1991, pp. 254, 270, n. 2), for instance, laments the functionalist’s
penchant for “boxology,” i.e. drawing diagrams that install component functions in
separate boxes, “while explicitly denying that these boxes have anatomical significance.”
While he concedes that “in principle” it may be a good tactic, and one which he himself
has employed, “it does tend to blind the functionalist to alternative decompositions of
function, and particularly to the prospect of [neural reuse].” Small wonder then that he
calls for a “better vision, anchored in a positive acceptance—as opposed to a hysterical
dismissal—of the foundational facts of functional neuroanatomy” (see also McGeer 2007;
Hardcastle & Stewart 2009, p. 194; Karmiloff-Smith 1994, p. 702; Bechtel 2008a, p. 990;
Gerrans 2014, pp. 22-23). In § 8.2 below I put the two primary empirical arguments in
favour of MR under pressure, as well as offer a survey of recent arguments skeptical of
the MR hypothesis.

But let me be clear at the outset. My position is in no way hostage to the fortunes
of MR, however one cares to define it. There are other ways to argue for the pertinence of
neuroscience to psychology that need not presuppose type identity among cognitive and
neural states. I can think of at least three. For one thing, it does not pay to have an
uncompromisingly rigid understanding of the reduction relation. John Bickle (2010, pp.
250-251; 1998, p. 30) appeals to examples of successful reduction from the history of
science that happened to involve MR kinds as among the posits of the reduced theories,
including examples of reduction involving radical MR—that over distinct physical states
occurring within the same token physical system at different times. The reduction of
temperature to mean molecular kinetic energy could be considered to involve radical MR,
since classical thermodynamic kinds like temperature are in truth macroscopic states
multiply realized over different microstates of the same macroscopic system over time.
And yet the reduction of classical thermodynamics to the kinetic/corpuscular theory of
matter is “the textbook example of scientific intertheoretic reduction.” In another vein,
Jaegwon Kim (1992), Larry Shapiro (2000) and Colin Klein (2008) have each drawn
attention to a significant dilemma confronting the MR advocate. If a given functional
kind is not multiply realized, the traditional argument for its autonomy and irreducibility

150
falls away. If on the other hand the kind is multiply realized, the kind will not be a proper
scientific kind, i.e. of the sort that can enter into laws. “Brittleness” might be a multiply
realized property, but glass, steel and biscuits are each brittle in their own way: there can
be no general science of brittle things. Ranging over such genuinely diverse physical
realizations means the kind will not enter into laws (i.e. exhibit lawlike or projectable
properties), except for those which are true analytically—such as all mousetraps catch
mice, and all eyes see—and this in turn “undercut[s] the traditional motivation for
admitting functional kinds into the ontologies of the special sciences” (Shapiro 2000, p.
637). Thus if mental states really are multiply realized, neuroscience will matter very
much indeed—not so much because psychology could not do without it, but because
psychology’s claim to be a traditional science would be open to question. This is only the
most skeptical conclusion one could draw, but the point is well-taken (see Couch 2009a,
pp. 262-264 for criticisms, however). Finally, and somewhat trivially, multiply realizable
does not mean infinitely realizable. Cognitive hypotheses will always have implications for
realizers—mousetraps cannot be made of paper after all. A functionalist psychology, this
is to say, can proceed only within the biophysical limits that its own constructs impose,
and ignores evidence of implementation to its peril. Take modularity as an example. A
commitment to modularity standardly entails a belief in the functional dissociability of at
least some cognitive capacities; but this, as we saw in Chapters 4, 5 and 7, is just not the
sort of feature which the evidence of reuse makes available, at any rate in a
straightforward manner. Of course neural hypotheses may themselves be disconfirmed by
evidence coming from other branches of the cognitive sciences, including psychology, so
the constraints here are genuinely bidirectional and intertheoretic. But MR or not, there
is simply no way of getting around the neuroscience (McGeer 2007).

Be that as it may, given the recent tide of empirical challenges to the MR thesis,
and because MR has proven itself to be an occasional stumblingblock in the path of those
committed to the autonomy of the special sciences, I have considered it worthwhile saying
at least something on the subject. While the more austere school of functionalism
admittedly no longer enjoys the following it once had—and mainstream functionalists
today would hardly dismiss neuroscience on the basis of psychology’s autonomy of
neuroscience—pockets of the austere school do survive under the guise of cognitive
neuropsychology and related fields (see McGeer 2007 for detailed analysis and criticism).

151
 8.2 MULTIPLE REALIZATION REVISITED

8.2.1 Preliminary remarks

Two primary empirical arguments have been advanced at one time or another in support
of the view that cognitive states are multiply realized. One argument proceeds from
evidence of extensive neuroplasticity in the brain. The other proceeds from an account of
convergent evolution. In the first part of this section I shall briefly address these
arguments for MR and the existence of MR kinds. In the second part I provide a
conspectus of the most forceful arguments against MR developed in recent years.

8.2.2 Empirical arguments for multiple realization

8.2.2.1 Neuroplasticity

Neuroplasticity has in recent times been thought to provide compelling evidence for the
MR of mental states. Shapiro (2004) and Polger (2009) review this evidence and find that
it does not provide evidence of MR. Polger (2009, p. 470) concludes that “contrary to
philosophical consensus, the identity theory does not blatantly fly in the face of what is
known about the correlations between psychological and neural processing.”

As we saw in Chapter 2, there is more than one kind of brain plasticity, including
inter alia cortical map plasticity and synaptic plasticity. Very roughly, the former occurs
when different brain regions subserve the same function at different times in an
individual’s history, say, after brain injury or trauma, and it is this plasticity which is
most often regarded as supporting MR. Synaptic plasticity refers to the strengthening or
weakening of connections between neurons, and is believed to have a role in learning and
memory (and quite possibly, therefore, in cortical map plasticity). I shall restrict myself to
the first kind here.

Evidence telling against type distinct kinds, and therefore supporting a modest
identity relation holding between psychology and neuroscience, includes the following
(Polger 2009, pp. 467-468).

152
 • Cortical maps do not migrate wildly, i.e. they do not simply “ ‘jump’ to recruit
unused but non-adjacent cortical areas,” and when they appear to do so it is
generally to exploit the structural features common to different sites (see e.g. the
discussion of EB in Chapter 6).
• Recovered functions are frequently suboptimal—genuine MR would require the
same psychological state to be underwritten by different neurological states;
suboptimality is evidence that the psychological states are in fact different, and
therefore evidence of difference underlying difference, not difference underlying
sameness, as MR requires. Actually this argument requires care. It only goes
through if it can be shown that there is a distinct function served by the two
psychological states. I address this matter in the next chapter.
• Functional studies of the rewired ferrets whose visual cortex was induced to
project into auditory cortex suggest they “developed processing structures—in
particular, columnar organization—that is [sic] typical of visual processing”
within auditory cortex. Since auditory cortex came to resemble visual cortex
(Shapiro 2008, p. 518), this is not evidence that auditory regions learned to handle
visual domain tasks while still in their auditory configuration, as genuine MR
would require here. After all, auditory regions have specific neural configurations
and connection patterns. If these are not preserved when the ferret’s auditory
cortex begins processing visual input, it cannot be a case of MR. What we have is
a case of sameness underlying sameness, not difference underlying sameness, as
MR requires. Put another way, the studies are reporting a genuine case of
crossmodal plasticity (or perhaps supramodal plasticity), not MR (see §§ 2.4.2-
2.4.3).

In fact it is unsurprising that neuroplasticity has not been able to deliver up the expected
argosy of empirical support for MR, and this for two reasons. Firstly, many of the most
sophisticated brain imaging techniques to date have not been able to yield high resolution
mappings of the neural configurations implicated in rehabilitation after injury. Neither
positron emission tomography (PET) nor fMRI measures neural activity and network
configuration directly. What they measure, in fact, is blood flow, which can hardly tell us
much about whether the functions in question are multiply realized. Even if MR were

153
pervasive, these methods would not yield coherent, interpretable results. Why then
assume that a recovered function with a new location should have a different realization
basis, unless of course there is some other indication that makes it likely (e.g.
simultaneous preservation of the old and recovered function in the new area)? Secondly,
without standardized criteria for evaluating sameness/difference judgments, how can we
be sure that neuroplasticity even speaks to MR? I conveniently neglected this
consideration when making the points above, and in fact it is only in recent years that
more careful attention has been paid to questions of sameness and difference in debates
about MR. I take this up in § 8.2.3.4.

8.2.2.2 Convergent evolution

Evolutionary considerations, particularly the idea that convergent evolution is likely to

generate psychological similarities as well as behavioural similarities (such as flight in
birds and bats) in morphologically unrelated species, have been thought to weigh in
favour of MR. But the issue cannot be decided a priori. MR is an empirical hypothesis in
the end and must sooner or later come to terms with empirical evidence. In fact,
evolutionary considerations might actually tell against MR. Here I cite only one case,
detailed at greater length in Bickle (2003; 2010). The process by which short-term
memory becomes long-term memory is known as memory consolidation, and Bickle cites
evidence supporting the likelihood of there being shared molecular mechanisms for
memory consolidation across biological taxa as diverse as fruit flies, sea slugs and mice.
One might not think this sort of evidence admits of any far-reaching consequences for
human psychology or MR generally, but if the instance is seen to follow from certain
“core principles of molecular evolution,” it assumes a larger significance. One such
principle holds that the amino acid sequences of specific proteins in functionally
important, “constrained,” domains change much more slowly than in functionally less
important domains. This principle implies the existence of “universally conserved”
molecular mechanisms across distantly related biological species, albeit those found deep
down in cellular physiology and intracellular signaling pathways, just as the
fly/slug/mouse pathway attests.

In the end, any psychological kind that affects an organism’s behavior must
involve the cell-metabolic machinery in individual neurons. In the brain, causally
154
 speaking, that’s where the rubber hits the road. But that’s the very machinery
[that] tends to be conserved evolutionarily across existing biological species.
Random changes to its components, especially to amino acid sequences in its
proteins’ constrained domains (almost) inevitably are detrimental to an organism’s
survival. (2010, p. 258)

Far from being exceptional, “molecular evolution suggests [such mechanisms] should be
the rule.” So at least this empirical argument, rooted in considerations of evolutionary
plausibility and molecular evolution, predicts that the MR thesis is false at the molecular
level, if not at the systems level (see also Hawrylycz 2015, pp. 8-9; Koch 2015, p. 26 and
Zador 2015, p. 43).

8.2.3 A conspectus of recent arguments against multiple realization

8.2.3.1 Outline of arguments

The most powerful arguments against the MR hypothesis as presented in the recent
literature include:

• the argument from comparative psychology (Bechtel & Mundale 1999);

• the argument from grains (Bechtel & Mundale 1999); and
• the argument from context (Bechtel & Mundale 1999; Shapiro 2000; Polger &
Shapiro 2008; Shapiro & Polger 2012).

8.2.3.2 The argument from comparative psychology

Bechtel and Mundale (1999) appeal to “neurobiological and cognitive neuroscience

practice” in the hope of showing how claims that psychological states are multiply
realized are unjustified. Essentially, theirs is an argument from success: cognitive
neuroscience’s method assumes MR is false, and the success of that method is evidence
that MR is false. They argue that it is “precisely on the basis of working assumptions
about commonalities in brains across individuals and species that neurobiologists and
cognitive neuroscientists have discovered clues to the information processing being
performed” (1999, p. 177).

155
 Bechtel and Mundale examine both the “neuroanatomical and neurophysiological
practice of carving up the brain.” What they believe this examination reveals is, firstly,
that the principle of psychological function plays an essential role in both disciplines, and
secondly, that “the cartographic project itself is frequently carried out comparatively—
across species” (1999, p. 177), the opposite of what one would expect if MR were “a
serious option.” It is the very similarity (or homology) of brain structures which permits
generalization across species; and similarity in the functional characterization of
homologous brain regions across species only makes sense if the claims of MR are either
false or greatly exaggerated. For instance, “[e]ven with the advent of neuroimaging,
permitting localization of processing areas in humans, research on brain visual areas
remains fundamentally dependent on monkey research…” (1999, p. 195). Brodmann’s
famous brain maps were based upon comparisons of altogether 55 species and 11 orders
of mammals. If MR were true, “one would not expect results based on comparative
neuroanatomical and neurophysiological studies to be particularly useful in developing
functional accounts of human psychological processing” (1999, p. 178). They also argue
that the ubiquity of brain mapping as a way of decomposing cognitive function points to
the implausibility of the MR thesis. The understanding of psychological function is
increasingly “being fostered by appeal to the brain and its organization” (1999, p. 191),
again, the opposite of what one would expect “[i]f the taxonomies of brain states and
psychological states were as independent of each other as the [MR] argument suggests”
(1999, pp. 190-191).

Aizawa (2009, pp. 500-503) detects a tacit claim in Bechtel and Mundale to the
effect that unique cross-species localization of functions in the brain entails their unique
realization. This is thought to be a non sequitur. It is true that, strictly speaking, what
much of their paper succeeds in showing is the unlikelihood of “multiple localization,”
but two things can be said in response. Firstly, the criticism to some extent misses the
point of their analysis. Bechtel and Mundale have deliberately opted for functional
localization, i.e. brain activity in the same parts or conglomerate of parts across species, as
the relevant standard by which to judge the sameness or difference of brain states, and
they have done so in deference to cognitive neuroscience practice. Localization is for
them the appropriate standard to adopt because it is at the right grain of description (see
next section). Secondly, it is not actually easy to police the distinction between

156
localization and realization, for neural localization is an important dimension of neural
organization. Aizawa complains that Bechtel and Mundale do not provide independent
evidence in support of any such proposition, but, all in all, it does not seem to be a
particularly tendentious one.1 It is true that, for all we know, functions could be localized
in the same region in closely related species, and yet be organized very differently. But
surely we would need some reason to think that this is likely.

A more serious criticism of their argument is that it speaks only to species that are
homologous—it might demonstrate that the ubiquity of MR is questionable so long as we
restrict our gaze to primates and rodents (i.e. mammals generally, related by common
descent), but surely it does not succeed in showing that octopuses and humans realize
their psychologies in anything like the same way (Kim 2002; Shapiro 2008). Nevertheless,
the arguments grounded in methodological and comparative considerations offer an
impressive refutation of MR among the many which have been attempted in the past
fifteen or so years. Clearly there is some notion of similarity at stake here which, while
largely unarticulated in Bechtel and Mundale, ultimately needs to be reckoned with—at
least among homologues (cf. Gillett 2002; 2003; Polger & Shapiro 2008).

Next I present two distinct but related arguments, the argument from grains and
the argument from context. Bechtel and Mundale have something to say about both,
whereas Shapiro confines his analysis, so far as we shall be concerned, to the import of
context alone.

8.2.3.3 The argument from grains

Bechtel and Mundale (1999, pp. 178-179, 201-204) resort to grains as a way of making
sense of what they perceive to be the entrenched, almost unquestioning consensus
prevailing around MR. They think this can be traced to the practice of philosophers
appealing to different grain sizes in the taxonomies of psychological and brain states,
“using a coarse grain in lumping together psychological states and a fine grain in splitting
brain states.” When Putnam went about collecting his various specimens of pain, he

1It only appears to be tendentious when a certain paradigm of realization and MR, the so-called
“dimensioned” view, has one under its sway (see Gillett 2003).

157
ignored the many likely nuances between them. At the same time, he had few
compunctions about declaring them different at a neurological level. His contention that
pain is likely to be an MR kind can only command our respect if we can be sure that when
he was comparing his specimens from a neurological point of view he was careful to apply
no less lenient a standard of differentiation than he applied when comparing his
specimens from a psychological point of view. Bechtel and Mundale maintain that when
“a common grain size is insisted on, as it is in scientific practice, the plausibility of
multiple realizability evaporates.” As their examples of neuroanatomical and
neurophysiological practice attest, scientists in these fields typically match a coarse-
grained conception of psychological states with an equally coarse-grained conception of
brain states. Despite the habit of philosophers individuating brain states in accordance
with physical and chemical criteria, a habit no doubt originating with Putnam, this is not
how neuroscientists characterize them. The notion of a brain state is “a philosopher’s
fiction” (1999, p. 177) given that the notion neuroscientists actually employ is much less
fine-grained, namely “activity in the same brain part or conglomerate of parts.”

While sympathetic to this argument, I think it really works in one direction

alone—namely, when refuting the MR advocate who has helped herself to a fine-grained
notion of a brain state, while assuming a coarse-grained notion of a mental state. In other
cases, I think the argument as put is quite problematic. For completeness, and because it
is a highly influential argument, I shall elaborate upon the argument’s strengths and
weaknesses in the following chapter.

8.2.3.4 The argument from context

A not unrelated factor which Bechtel and Mundale think might help explain the
prevailing MR consensus in philosophy and the mind sciences is that the hypothesis itself
is often presented in a “contextual vacuum.” The choice of grain is always determined by
context, with “different contexts for constructing taxonomies” resulting in “different
grain sizes for both psychology and neuroscience.” The development of evolutionary
perspectives, for instance, in which the researcher necessarily adopts a coarse grain,
contrasts with the much finer grain that will be appropriate when assessing differences
among conspecifics. “One can adopt either a coarse or a fine grain, but as long as one uses

158
a comparable grain on both the brain and mind side, the mapping between them will be
correspondingly systematic.”

Shapiro (2000), more alive than most to the acuteness of the metaphysical
problem that is central to MR, has provided perhaps the most useful and philosophically
perceptive treatment of sameness/difference judgments as they touch upon MR. He
states:

Before it is possible to evaluate the force of [the MR thesis] in arguments against

reductionism, we must be in a position to say with assurance what the satisfaction
conditions for [the MR thesis] actually are. (2000, p. 636)

For him:

The general lesson is this. Showing that a kind is multiply realizable, or that two
realizations of a kind are in fact distinct, requires some work. (2000, p. 645)

And:

To establish [the MR thesis], one must show that the differences among
purported realizations are causally relevant differences. (2000, p. 646)

Shapiro’s concerns revolve around what motivates ascriptions of difference, and therefore
sameness. The issue is important because the classic intuition pump that asks us to
conceive a mind in which every neuron has been replaced by a silicon chip depends on
our ascription of an interesting difference between neurons and silicon chips, apparently
even where silicon chips can be made that contribute to psychological capacity by one and
the same process of electrical transmission. His answer too, like Bechtel and Mundale’s,
depends ultimately on context—in particular, the context set by the very inquiry into MR
itself.

Shapiro (2000, pp. 643-644) argues that “the things for which [the MR thesis] has
a chance of being true” are all “defined by reference to their purpose or capacity or
contribution to some end.” This is the reason why carburetors, mousetraps, computers
and minds are standard fare in the literature of MR. They are defined “in virtue of what
they do,” unlike, say, water, which is typically defined by what it is, i.e. its constitution or
159
molecular structure, and accordingly not an MR kind. Genuine MR requires that there
be “different ways to bring about the function that defines the kind.” Truly distinct
(indeed multiple) realizations are those that “differ in causally relevant properties—in
properties that make a difference to how [the realizations] contribute to the capacity
under investigation.” Two corkscrews differing only in colour are not distinct realizations
of a corkscrew, because colour “makes no difference to their performance as a corkscrew.”
Similarly, the difference between steel and aluminium is not enough to make two
corkscrews that are alike in all other respects two different realizations of a corkscrew
“because, relative to the properties that make them suitable for removing corks, they are
identical.” In this instance, differences of composition can be “screened off.” Naturally
there may be cases where differences of composition will be causally relevant—perhaps
rigidity is the allegedly MR kind in question, in which case compositional differences will
necessarily speak to how aluminium and steel achieve this disposition. Each case must
simply be judged on its own merits, indeed in its own context (as Bechtel and Mundale
might put it). Thus unlike the two corkscrews identical in all respects save colour, which
do not count as distinct realizations, waiter’s corkscrews and winged corkscrews are
enabled to perform the same task in virtue of different causally relevant properties, and
therefore do count as genuinely distinct realizations of a corkscrew, one based on the
principle of simple leverage, the other relying on a rack and pinions (see Fig. 1).

(a) (b)

Figure 1. A waiter’s corkscrew (a) and a winged corkscrew (b). Each contributes to the capacity of cork-
removal in different ways.

The problem for the MR advocate is this: it is apparent that each of the examples
just given involves a straightforward mechanism that renders its classification reasonably
intuitive (though I dare say still subject to the odd disagreement!). With the brain the
situation is inestimably more complicated, so why should the MR advocate feel so
160
confident that there is more than one truly distinct way to build a mind, given that we
know comparatively little about how brains produce minds in the first place? One might
even think that the brain’s functional properties would need to be understood at least as
well as the corkscrew’s before one could venture an opinion about whether truly different
brains can produce similar minds.2

I shall have more to say about Shapiro’s analysis in the next chapter, because I
think it gets to the heart of the problem facing the functionalist. It is too often assumed
that MR is ubiquitous, and that the onus is on the identity theorist to prove otherwise.
Shapiro’s analysis explains why establishing that a kind is an MR kind actually takes
work.

8.2.4 Conclusion

The primary empirical arguments for MR resting on weak evidence at best, the
functionalist has more work to do before she can, by her lights with a clean conscience,
look askance at the neural evidence and its implications for cognitive theorizing.

8.3 NEW LOOK FACULTY PSYCHOLOGY

Exactly how the science of mental faculties will have to change to accommodate the
reality of neural reuse is a matter of some uncertainty, but even those such as Michael
Silberstein who assert that “the autonomy and irreducibility of folk psychology are
assured” concede that reuse means “scientific psychology must be heavily revised” (2016,
pp. 27-28). My position, and the position which it was the aim of the previous section to
make feasible, is that psychology and neuroscience are friends not enemies. I start from
the premise that the best way to understand something is to break it down, and that the
best and most natural way to break something down is to carve it at its joints, i.e. in such a
way as to respect its physical constitution and design. I take this to be near truism. Now
the fact remains that, as I have tried to show, the endeavour to understand the mind has
come a long way from the days when Herbert Simon and David Marr reasoned from

2Shapiro and Polger (2012, p. 282) elaborate upon Shapiro’s (2000) pragmatic considerations and attempt
to situate his criteria within a somewhat more formal rubric. See also Shapiro (2008, pp. 522-525); Polger
(2008); Polger (2009, pp. 463-464) and Polger and Shapiro (2016).

161
evolutionary principles that the carve-up of relevance to the mind produces
independently modifiable subcomponents that correspond to functionally specific units of
high-level psychology (Sternberg 2011, p. 158). I am not saying that the mind is bereft of
dissociable subcomponents (see Chapters 4 and 5), but our ideas about them have
certainly changed. So long as psychology wishes to carve nature at its joints, then, it will
have to update its ideas about what those joints are. At a minimum, neural reuse
mandates an approach to decomposition which assigns domain-neutral functional
assignments to brain regions; and if such regions are the stuff of higher level cognition, it
is surely not reasonable to insulate the higher level cognitive ontology from their effects.
One does not have to be a ruthless reductionist or eliminativist to recommend that our
sciences so develop as to facilitate mutual interaction and even potentially unification. If
this is right, cognitive models of distinct domains should be placed upon such a footing as
will best accommodate the possibility of interaction. We will learn more about the
faculties, not less, if we can appreciate their deeper level associations. It is interesting to
note in this respect that the 2010 edition of David Marr’s Vision—a book which is
(in)famous for having made the strict independence between levels of inquiry an article of
faith in the cognitive sciences—contains an afterword that chimes nicely with the message
I am trying to convey here:

(1) insights gained on higher levels help us ask the right questions and do the right
experiments at the lower levels, and (2) it is necessary to study nervous systems at
all levels simultaneously. From this perspective, the importance of coupling
experimental and theoretical work in the neurosciences follows directly; without
close interaction with experiments, theory is very likely to be sterile. (2010 [1982],
p. 364)

One example of this mutual endeavour can be seen in neurolinguistic work that
tries to integrate formal results from the Minimalist Program in syntax. (In adverting to
this work I do not mean to endorse the Minimalist Program, only to illustrate how
researchers in fields that have typically been seen as antagonistic to one another can come
together in the interests of science.) David Poeppel (2015) remarks that the goals of
systems neuroscience and research in syntax have aligned in the past two decades. I
touched upon some of the relevant systems neuroscience in my discussion of canonical
neural computations (§ 7.3). The discovery of primitive computations such as filtering,
divisive normalization and predictive coding bodes well for the basic assumptions behind

162
the Minimalist Program. Both research programs seek to uncover fundamental and (as
much as possible) domain-general operations underlying various cognitive phenomena.
Merge can be seen as another such computation (§ 7.2). That part of Merge which
combines elements is a close analogue of binding (or concatenation) operations within
systems neuroscience (Poeppel 2015, p. 144). Given the expressions A and B, the binding
operation produces a new expression (A, B à{A,B}). A separate procedure then labels
the output. This work is encouraging in one important respect. Traditionally a major
obstacle to collaboration between neuroscientists and linguists was the abstruseness and
intractability of transformational-generative rules. Simplification of these rules at least
makes interdisciplinary collaboration possible, and should be seen as a step in the right
direction inasmuch as many linguists now seem to have an eye to fundamental neural
computations.

Returning to reuse, as I suggested, the most straightforward outcome on the table

is for the higher level ontology to incorporate the lower level one—i.e. the level of
fundamental computations performed in modules and discrete brain regions. But what is
the nature of these primitives? Are we talking about discrete domain-general
computations in specific cortical sites, with a one-to-one mapping between primitives and
brain regions (in the manner contended by Russell Poldrack)? Or are we talking more
about dispositions, so that an individual brain region represents a particular complex of
primitives, with a many-to-many mapping between primitives and brain regions (in the
manner contended by Michael Anderson)? Then there is the issue of how faculty
psychology is to proceed in light of nondecomposition and network dynamics (see
discussion in § 5.1). Some structures (maybe many?), as we saw, are not classically
decomposable—their properties are not additive in a bottom-up sort of fashion—even
though functional decomposition is virtually an article of faith in cognitive science. These
are questions to be clarified and hopefully resolved in coming years. What they provide is
a sense of the terms that any negotiated settlement between psychology and neuroscience
will likely take, since it is certain that faculty psychology will have to reckon with these
primitives one way or another.

163
 8.4 SUMMARY

Multiple realization should not be taken as an empirical given—establishing that a kind is

multiply realizable takes a good deal of work, as Shapiro has been at pains to show; and
even when the existence of an MR kind can be verified, the details of its implementation
do not suddenly become irrelevant. Structure and function are two sides of the same coin.
Thus the multiple realization argument provides no basis for neglecting the discoveries of
neuroscience. Faculty psychology’s strength lies precisely in its willingness to work with
neuroscience.

164
 9
_____

Multiple realization and the

commensurability of taxonomies: A note on grains1

9.1 PRELIMINARY REMARKS

The past two decades have witnessed a revival of interest in MR. Bechtel and Mundale’s
(1999) illuminating discussion of the subject must no doubt be credited with having
generated much of this renewed interest. Among other virtues, their paper expresses what
seems to be an important insight about multiple realization: that unless we keep a
consistent grain across realized and realizing kinds, claims alleging the multiple
realization of psychological kinds are vulnerable to refutation. I introduced this “grains”
argument in the previous chapter. For completeness, the present chapter elaborates upon
what this argument gets right and what it gets wrong. The argument is certainly intuitive,
but intuitions notwithstanding, the terms of Bechtel and Mundale’s recommendation
make it impossible to follow, while also misleadingly insinuating that its application
virtually guarantees mind-brain identity. Instead of a matching of grains, what multiple
realization really requires is a principled method for adjudicating upon differences
between tokens. Shapiro’s (2000) work on multiple realization, which I also presented in
the previous chapter, can be understood as an attempt to adumbrate just such a method.
Indeed I argue that Shapiro’s work represents an ideal approach to refuting claims of MR.
While his “causal relevance” criterion can easily be mistaken for Bechtel and Mundale’s
grain requirement, my analysis reveals exactly where and why the two tests diverge.

1The material in this chapter was presented to the American Philosophy of Science Association on 3
November 2016, in Atlanta, Georgia. I am particularly indebted to Larry Shapiro and Tom Polger for
helpful discussion. A version of this chapter is due to appear in the journal Synthese (see Zerilli 2017b).

165
 9.2 RECEPTION OF THE GRAINS ARGUMENT

Recall that the grains argument alleges that MR advocates frequently employ a mismatch
of grains when characterizing mental and neural states. Bechtel and Mundale also think
that the MR hypothesis often gets presented in a “contextual vacuum.” The choice of
grain is always determined by context, with “different contexts for constructing
taxonomies” resulting in “different grain sizes for both psychology and neuroscience.” As
they explain:

One can adopt either a coarse or a fine grain, but as long as one uses a comparable
grain on both the brain and mind side, the mapping between them will be
correspondingly systematic. For example, one can adopt a relatively coarse grain,
equating psychological states over different individuals or across species. If one
employs the same grain, though, one will equate activity in brain areas across
species, and one-to-one mapping is preserved (though perhaps further taxonomic
refinement and/or delineation may be required). Conversely, one can adopt a very
fine grain, and differentiate psychological states between individuals, or even in
the same individual over time. If one similarly adopts a fine grain in analyzing the
brain, then one is likely to map the psychological differences onto brain
differences, and brain differences onto psychological differences. (1999, p. 202)

At least among some philosophers Bechtel and Mundale’s message has evidently
been well received (Couch 2004; Polger 2009; Godfrey-Smith, personal communication;
see also tacit approval in Aizawa & Gillett 2009, p. 573). Polger (2009) explains the
motivation for the grain requirement in an illuminating way. As I mentioned in the
previous chapter, neuroplasticity has often been supposed to provide compelling evidence
for the MR of mental states. Polger disagrees (in my view correctly). But the grains
argument figures prominently in his reasoning. As he points out, it might be tempting to
regard a phenomenon like cortical map plasticity—where different brain regions subserve
the same function at different times in an individual’s history, say, after brain injury or
trauma—as an existence proof of MR. But not if the point about grains is taken to heart.
It all comes down to what we mean by “different brain regions” subserving “the same
function.” As I noted earlier, recovered functions in these circumstances are frequently
suboptimal. Genuine MR requires the same psychological state to be underwritten by
different neurological states; but suboptimality is evidence of difference underlying
difference, not difference underlying sameness, as MR requires:

166
 It’s true that this kind of representational plasticity involves the “same” function
being mediated by “different” cortical areas. But here one faces the challenge
leveled by Bechtel and Mundale’s charge that defenses of [MR] employ a
mismatch in the granularity of psychological and neuroscientific kinds. If we
individuate psychological processes quite coarsely—by gross function, say—then
we can say that functions or psychological states are of the same kind through
plastic change over time. And if we individuate neuroscientific kinds quite
finely—by precise cortical location, or particular neurons—then we can say that
cortical map plasticity involves different neuronal kinds. But this is clearly a
mug’s game. What we want to know is not whether there is some way or other of
counting mental states and brain states that can be used to distinguish them—no
doubt there are many. The question is whether the sciences of psychology and
neuroscience give us any way of registering the two taxonomic systems. (2009, p.
467, my emphasis)

9.3 PROBLEMS WITH THE GRAINS ARGUMENT

But now the question is this: what, precisely, can it mean to use a “comparable” grain, or
to keep a grain size “constant,” across both psychological and neurophysiological
taxonomies? Polger’s motivation makes a lot of sense, to be sure, but talk of “registering”
taxonomies (as of aligning classificatory regimes, or rendering distinct scientific
descriptions commensurable, or however else one might care to put it) doesn’t shed any
light on how the desideratum for consistent grains can actually be met. Since it is
intended to serve in part as a methodological prescription, it’s important to know what to
make of this requirement—metaphors won’t help us here. How, in concrete terms, is an
investigator meant to satisfy such a condition as this on their research?

Let me begin by using pain and hunger as the MR kinds in question, both because
they are states about which Bechtel and Mundale have something to say and because they
have long been staples of the MR literature. In the previous section I mentioned that
when Putnam went about collecting his various specimens of pain, he ignored the many
likely subtle differences between them—even as he had few scruples about declaring
them different at a neurological level. Bechtel and Mundale’s complaint against Putnam
therefore seems to be that when he says that pain (or hunger) is likely to be an MR kind,
we can only go along with him if we can be sure that, when he was comparing his
specimens from a neurological point of view, he was careful to apply no less lenient a
standard of differentiation than he applied when comparing his specimens from a

167
psychological point of view. But because in their view he didn’t do this, he opened
himself to their challenge that he was unduly permissive in his classification of
psychological states. This explains why Bechtel and Mundale chide that:

A human’s psychological state and that of an octopus might well be counted as the
same insofar as they are associated with some general feature (such as food-
seeking behavior, in the case of hunger). But with respect to other considerations,
a human psychological state may be considered different from that of an octopus,
even if we limit the scope to mere behavior. Food-seeking behavior for the
octopus is different from food-seeking behavior in the human if one is concerned
about such things as how one seeks the food, what foods are sought, under what
conditions, etc. (1999, p. 203)

This seems intuitive and sensible. In the present context, therefore, the word “grain”
seems to mean something like, “the basis for comparison,” or more precisely, “the respect
under which we seek to compare a set of tokens”; and “same (grain)” means something
like “equally permissive or stringent (bases of comparison).” I think this takes us a step
closer to what Bechtel and Mundale mean when they exhort us to employ the same grain
across realized and realizing kinds. MR requires a comparison of a set of tokens from both
a psychological and a neurological point of view; and we must see to it (somehow) that
these cross-disciplinary reference points are on talking terms. So far so good.

But now what does this mean? How can the bases for comparing a set of tokens (in
the present context, two brains) be “on talking terms” or “commensurable” or
“comparable” when the bases are delivered by two distinct disciplines (in the present
context, psychology and neuroscience)? I’m going to illustrate what I take to be the most
natural way in which such cross-disciplinary reference points can be brought into
alignment by using an example where the two disciplines are more clearly distinct than
psychology and neuroscience. In this way the nature of the problem will be brought into
much sharper relief than if I were to stick with psychology and neuroscience. Suppose
you have two tokens of fruit. The science of botany (say) could deliver descriptions under
which the two are classified the same (e.g. from the point of view of species), but also
descriptions under which they come out different (e.g. from the point of view
of varieties). The first description could be said to apply a coarser grain than the second.
Now imagine economics coming into the picture. The science of economics can likewise
deliver descriptions under which both tokens are classified the same (e.g. both are forms

168
of tradable fresh produce) or different (e.g. one, being typically the crunchier and sweeter
variety, has a lower elasticity of demand than the other). Once again, the first description
could be said to apply a coarser grain than the second. Perhaps, then, we could take it that
botany and economics deliver descriptions at the same grain of analysis when their
judgments of sameness or difference cohere in a given case. In the example, botanical
descriptions via species classification would be furnished at the same grain as economic
descriptions via commodity classification, so that species descriptions in botany are “at
the same grain” as commodity descriptions in economics. By the same logic, variety
descriptions in botany would be comparable to elasticity descriptions in economics.

This construal is fairly reasonable, I think, and it’s a serious, charitable attempt to
make sense of Bechtel and Mundale’s recommendation. Still, there’s a big problem
here. If this is all that “maintain a comparable grain” amounts to, it really does beg the
question, for this is simply type-type identity by fiat. Of course such a recommendation
will ensure that the mapping between psychology and neuroscience will be “systematic”
(to use Bechtel and Mundale’s term), because on this account yielding concordant
judgments of similarity or difference across taxonomies is simply what it means to apply
the same grain. So this version of the grain requirement makes type-type identity a fait
accompli, effectively obliterating all MR kinds from the natural order.

It’s just as well that I don’t think this is quite what Bechtel and Mundale had in
mind when they made their move to grains. And yet they do say: “One can adopt either a
coarse or a fine grain, but as long as one uses a comparable grain on both the brain and
mind side, the mapping between them will be correspondingly systematic” (note that—
it will be!). This sounds like someone with the utmost confidence in the grain
requirement, which is of course what one would have if one thought grains could be
legitimately matched in just this way. In the same passage they assert that, in the context
of a researcher invoking a relatively coarse grain to equate psychological states across
different individuals or species, “[i]f one employs the same grain…one will equate
activity in brain areas across species, and one-to-one mapping is preserved…” (my
emphasis). “One will equate…”? How can they be so sure? My guess is that, while they
do have something important to tell us about MR, a beguiling metaphor has led them to

169
suppose that MR is easier to refute than it actually is. (I’ll support this contention with a
few examples in a moment.)

If not by this means, then, how else might two taxonomic descriptions be
rendered commensurable? Perhaps when I suggested a moment ago that two sciences
deliver descriptions at the same grain of analysis when their judgments of sameness or
difference cohere in a given case I should have added “likely (to cohere).” We would then
have:

Any two sciences deliver descriptions at the same grain of analysis when their
judgments of sameness or difference are likely to cohere in a given case.

This would at least solve the problem of inevitability, and makes sense of Bechtel and
Mundale’s more cautious choice of words near the same passage I cited earlier, where
they say, in the context of a researcher invoking a fine grain to differentiate psychological
states, “[i]f one similarly adopts a fine grain in analyzing the brain, then one is likely to
map the psychological differences onto brain differences, and brain differences onto
psychological differences” (1999, p. 202, my emphasis). “Is likely to” is better than
“will.” Still, this more cautious formulation doesn’t immunize them from the serious
errors I point out below. It appears that the distinction between likelihood and
inevitability must be handled very carefully in this debate, as is shown by Bechtel and
Mundale’s following remarks on computer programs:

One often speaks of running the same program on different computer

architectures, thereby generating a case of multiple realization. But in fact one has
surreptitiously adopted a coarse grain in characterizing the program. If one looks
carefully enough at performance measures, one will generally be able to identify
differences in the way the program runs on different computers.” (1999, pp. 202-
203, my emphasis)

How carefully need we look before we can stop looking? Until the mapping is one-to-one
between software and hardware? What if we don’t find a one-to-one mapping, having
already looked more carefully at the software, but could have a one-to-one mapping if
only we look yet a little more carefully? Do we stop, or keep going? How do we know?
The road from a likely one-to-one mapping to an inevitable one-to-one mapping is here at
its most precarious.

170
 But let’s try to remedy this if we can. Bechtel and Mundale are obviously after
some feature of taxonomic description that can, as it were, latch onto a corresponding
feature of another taxonomic description. Perhaps this makes most sense if we replace the
idea of a “grain” with the idea of a “level,” and more specifically a “level of description or
abstraction.” A software program may be described at different levels (machine language,
assembly language, compiler language, etc.), as indeed may its underlying hardware
(microphysical events, transistors, chips/circuits, etc.). This at once provides a much
clearer picture of how we might go about scoring off levels from one discipline against
another (and thereby achieve that most elusive and sought-after match between grains).
Higher levels of software will correspond to higher levels of hardware—and it need not be
inevitable that two programs with the same compiler language will be the same at the
circuit level, even though the compiler language occupies (let’s say) the corresponding
rank within the software hierarchy that the circuit level does within the hardware one.
Translating this thought into the present context, different psychological levels must be
understood as tallying up with different neurological levels. When the levels of
description correspond, the descriptions are at the “same grain.”

This does seem more promising, I agree, but I suspect that it is beset with more
problems in turn. For example, can we really lay out these two sciences (psychology and
neuroscience—indeed any two sciences!) on the Procrustean bed of “comparable levels”?
And more worryingly, who gets to say whether a level within the hierarchy of one
discipline really tallies with a level drawn from the hierarchy of another discipline? I fear
that replacing “grains” talk with “levels” talk serves only to demystify temporarily. The
problem has been merely postponed.

Of course matters aren’t much helped by the reasonable suspicion that MR is the
result of pairing inconsistent grains. For what is neuroscience if not a fine-grained
description of psychology, and psychology if not a coarse-grained description of
neuroscience? It is surely plausible that the neural and psychological sciences line up in
something like this way, given that talk about the mind is really talk about the brain from
a somewhat more abstract point of view.

171
 What Bechtel and Mundale are ultimately trying to convey through their
discussion of grains is the thought that claims of MR cannot be advanced willy-nilly—
that there is an objective and standard way to go about verifying the existence of MR
kinds and arbitrating disputes involving them. For the reasons just canvassed, however, it
strikes me that talk of grains doesn’t serve their purposes at all well. In fact they would
have been nearer the mark had they said that what MR requires is some sort of principled
mismatching of grains.

So far I’ve tried to indicate in what respects Bechtel and Mundale’s grain
requirement is imprecise and impracticable. Before I can show that the grains strategy is
also misleading, and actually often gives the wrong answer, it will be useful to set it
against an account which demonstrably gets things right, indeed an account which even
its detractors concede gets something very important right (e.g. see Gillett 2003, pp. 591,
592, 596, 597, 599, 600). Shapiro’s (2000) analysis, which we met with in the previous
chapter, expresses with enviable lucidity what I think is the crucial insight towards which
Bechtel and Mundale were uneasily groping. Interestingly, some philosophers—e.g.
Polger (2009)—write as if the grain requirement and Shapiro’s own formula for MR were
effectively interchangeable. This is a mistake: the two approaches deliver different
judgments in nontrivial cases (as I’ll illustrate in a moment).

Both Bechtel and Mundale and Shapiro would agree that the context of an inquiry
is relevant to the question that a researcher asks (indeed this is trivially true). The crucial
insight behind Shapiro’s analysis is that MR itself (at least partly) sets the context of the
inquiry here, inasmuch as MR makes function the primary basis of comparison between
tokens, i.e. the specific point of view from which we will compare a set of tokens in the
first instance (not phenomenology, not behavioural ecology, or anything else for that
matter). MR is after all a thesis about sameness and difference (Polger 2009); and any two
particulars will both differ and resemble infinitely (Bechtel & Mundale 1999, p. 203).
Whether two particulars are “the same” or “different” therefore depends on the aspect
from which we choose to compare them. It just so happens that to ask whether a kind is
an MR kind is already to have decided this preliminary question: what we want to know
here is whether two tokens that serve a particular function do so in the same way.
Explanatory considerations may of course fine-tune the sort of function that captures our

172
attention (cork-removal, rigidity, vision, camera vision, etc.). But function here is our key
preoccupation, and having settled on a specific function which a set of tokens can be said
to perform, the all-important question on Shapiro’s analysis is how the two tokens bring
that function about. Each case must be judged on its own merits.

Notice that to the extent Shapiro’s causal relevance criterion envisages certain
realizing properties being “screened off” from consideration in the course of inquiry,
there is a sense in which the taxonomies of realized and realizing kinds may be said to be
“commensurable” or “registrable” (no doubt explaining why some philosophers have
simply confused commensurability with causal relevance). Thus when comparing the
cork-removing properties of two waiter’s corkscrews, compositional differences will not
feature in the realizing taxonomy (if we accept Shapiro’s characterization of the problem).
So we have cork-removal, which features in what we may regard as a coarse-grained
taxonomy, realized by two objects described by a “science” of cork-removal in which
microstructural variations do not matter, hence which might also be regarded as a coarse-
grained taxonomy. If on the other hand we were comparing the same corkscrews for
rigidity, where one was made of steel and the other of aluminium, compositional
differences would feature in the realizing taxonomy. Here we would have rigidity, which
features in what we could well regard as a more fine-grained taxonomy than that
encompassing cork-removal, realized by two objects described by a science in which
microstructural variations really do matter (namely metallurgy), and which might also be
regarded as a fine-grained taxonomy, at least more fine-grained than the fictitious science
of cork-removal. But my point is this: commensurability nowhere appears as an
independent criterion of validity in Shapiro’s account of MR, for it is an artifact of the
causal relevance criterion, not a self-standing principle. Taxonomic commensurability is
in fact an implicit requirement of the causal relevance criterion in the sense that it’s taken
care of once the proper question is posed. As an explicit constraint it is a will-o’-the-wisp.

Armed with this analysis, let’s examine how Bechtel and Mundale attempt to
refute the status of hunger as an MR kind. Putnam (1967) had compared hunger across
species as diverse as humans and octopuses to illustrate the likelihood that some
psychological predicates are multiply realizable. On the basis of their grains critique,
however, Bechtel and Mundale suggest that hunger will not do the work Putnam had cut

173
out for it; for “at anything less than a very abstract level,” hunger is different in octopuses
and humans (1999, p. 202). The thought is that a finer individuation of hunger refutes the
existence of a single psychological kind, hunger, which can be said to cross-classify
humans and octopuses. Thus they essay to challenge the cognitive uniformity which MR
requires at the level of psychology.

Perhaps we might first note that when identifying a single psychological state to
establish the necessary conditions for MR, nothing Bechtel and Mundale say actually
precludes the choice to go abstract. If context is what fixes the choice of grain (as they are
surely right to point out), who’s to say that context couldn’t fix the sort of grain that
makes hunger relevant in an abstract sense? It may be tempting to think that a more
detailed description of something is somehow more real. But there is of course nothing
intrinsically more or less real about a chosen schema relative to others that might have
been chosen (there is no reason to suspect, for instance, that a determinate has any more
reality than a determinable). This consideration applies with no less force when we decide
to let the sciences dictate what the relevant kinds will be, since different sciences can take
both more and less abstract objects within the ken of their inquiries (Craver 2007). In fact
one way of reading Bechtel and Mundale’s paper (particularly §§ 2-4) is precisely as
suggesting that kind individuation should be fixed by the sciences. This point comes
through pretty clearly when they discuss the individuation of neural states. In that
specific context their paper provides excellent and detailed examples of how scientific
taxonomies can be invoked in debates over MR. The philosophical notion of a brain state,
they complain, is a “philosopher’s fiction,” because brain scientists by and large do not
individuate brain states on the basis of physical and chemical criteria; it is enough for
their purposes to employ the much coarser notion of “activity in the same brain part or
conglomerate of parts.” This is a clear demonstration of how regard for the sciences may
frequently mean opting for less precise/more abstract descriptions of phenomena.2

Notice, incidentally, that when it comes to the individuation of psychological

states (as distinct from neural states), Bechtel and Mundale’s advice about heeding the
sciences seems to get lost. The part of their paper where they do most to explicate a fine-

2Bechtel and Mundale’s test at this point can be seen as a bowdlerized version of Shapiro’s causal relevance
criterion, which screens off causally irrelevant details when comparing realizing kinds (much as Bechtel and
Mundale’s “coarse-graining” procedure does).

174
graining procedure for psychological kinds (1999, pp. 201-204) nowhere exemplifies how
the cognitive sciences can be invoked to guide individuation of psychological kinds.3
Instead they rely on “kind splitting” (Polger & Shapiro 2016), where a given (higher
level) kind (e.g. vision, as instanced in both humans and molluscs) is described in
increasingly greater detail until the illusion that two tokens are the same is finally
dispelled (allegedly demonstrating, for example, that human eyes and mollusc eyes are in
fact different psychological kinds, not instances of the same overarching psychological
kind “eye”). But kind splitting is gratuitous if it ignores the explanatory salience of the
higher level kind (see below), and in none of their examples (see 1999, pp. 201-204) are
the cognitive sciences actually consulted to check for this salience. This is precisely the
point at which the cognitive sciences would be most instructive: we need to be sure
whether a split is justified, and thus whether a split would really undermine the integrity
of a given higher level kind. So at least in the context of individuating psychological
kinds, it would appear that Bechtel and Mundale have failed to heed their own advice.

And yet there is a deeper problem with Bechtel and Mundale’s deployment of the
grains strategy here. To repeat their complaint: “at anything less than a very abstract
level,” hunger is different in octopuses and humans. But now why should this be
relevant? Who would deny it? They themselves seem to be oblivious to the context which
the very inquiry into MR makes paramount. They are not right to allege, as they do, that
“the assertion that what we broadly call ‘hunger’ is the same psychological state when
instanced in humans and octopi has apparently been widely and easily accepted without
specifying the context for judging sameness” (1999, p. 203). The reason why hunger,
pain, vision and so on were all taken for granted—assumed to be uniform at the cognitive
level—is because MR made function the point of view from which tokens were to be
compared. As Shapiro reminds us, “the things for which [the MR thesis] has a chance of
being true” are all “defined by reference to their purpose or capacity or contribution to
some end.” It was understood that, say in the case of pain, regardless of phenomenal,
ecological or behavioural differences between human and octopus pain (I doubt any of
which were lost on Putnam), all instances of pain in these creatures had something like
detection and avoidance in common. This might be to cast pain at “a very abstract level,”

3The nearest they come is this: “When comparing psychological states across different individuals,
psychologists…tend to ignore differences and focus on commonalities” (1999, p. 202). If anything,
however, one would have thought that this fact should actually discourage kind splitting (see text).

175
but this just happens to be the context which the inquiry into MR itself sets. A similarly
abstract feature is what unites all instances of hunger: let’s call it nutrition-induction. It is
not that decades of philosophers had simply forgotten to specify the point of view from
which these psychological predicates were being considered: it is rather that they simply
didn’t need to, since all of them had read enough of Putnam and the early functionalists
to know what they were about. Phenomenal and other differences that one might care to
enumerate between these predicates come a dime a dozen. But the whole point of
functionalism was to abjure the inquiry into essences and focus instead on the causal role
of a mental state within the life of an organism. Yes, this is to compare tokens from an
“abstract level,” but that’s what made functionalism intriguing to begin with. And if
Shapiro’s analysis is any guide, it is really the next step in the endeavour to verify the
existence of an MR kind that is the crucial one. Genuine MR requires that there be
“different ways to bring about the function that defines the kind.” So the follow-up
question concerns how the relevant organisms achieve their detection and avoidance
function, or nutrition-induction function, or whatever the case may be. It is in fact only
by asking this next question that we can appreciate just how badly the grains strategy
fares. The attempt to individuate hunger more finely would not refute the multiple
realizability of hunger as between humans and octopuses if it could ultimately be shown
that humans and octopuses achieve their nutrition-induction capacities in different ways
(which seems extremely likely); for then the functional role of hunger would be played by
two different realizers—the issue to which the MR inquiry is directed after all—
notwithstanding that the functional role can be more finely described within a more
specific frame of reference. Similarly, the attempt to individuate pain more finely would
not refute the multiple realizability of pain as between humans and octopuses if it could
ultimately be shown that humans and octopuses achieve their detection and avoidance
capacities in different ways (which again seems extremely likely). The sort of type
reduction that cuts its teeth by fine-graining psychological categories (i.e. kind splitting)
is almost always trivially possible, since at the limit everything is a special case; but the
move won’t succeed in refuting the existence of an MR kind at a higher level if the higher
level functional kind can be brought about in causally distinct ways. Of course there are
occasions when kind splitting may be mandated by the sciences, because the higher level
kind has no taxonomic relevance. This makes sense if kind individuation should be fixed
by the sciences (Couch 2009a; 2009b). In such cases the split really would undermine MR

176
at the higher level. But it’s important to emphasize that these cases require careful
consideration and may turn out to be rare in any event (see Polger & Shapiro 2016, pp.
103, 104-105, 110-111; Couch 2009a, pp. 262-264, 267; 2009b, p. 514).

So we see that the grains strategy, to the extent that it involves fine-graining
psychological states in order to undermine the cognitive uniformity required by MR, sets
itself a very easy job indeed, and mischaracterizes the nature of MR by its neglect of
function. Moreover Shapiro’s causal relevance criterion—which honors the core concerns
motivating Bechtel and Mundale’s resort to grains—does not demonstrate that hunger (or
pain) is type-reducible.

A good illustration of the grains strategy in action is provided by Couch’s (2004)

attempt to refute the claim that the human eye and the octopus eye are distinct
realizations of the kind eye. Conceding differences at a neurobiological level, the strategy
again involves challenging the alleged uniformity at the cognitive level. As he explains,
“[e]stablishing [MR] requires showing that…the physical state types in question are
distinct [and] that the relevant functional properties are type identical. Claims about
[MR] can be challenged at either step” (2004, p. 202). Reminding us that psychological
states “are often only superficially similar,” and that “at a detailed level the neural
differences make for functional differences” (2004, p. 203), he states:

Psychologists sometimes talk about humans and species like octopi sharing the
same psychological states. However, they also recognize that there are important
differences involved depending on how finely one identifies the relevant
features...Establishing multiple realization requires showing that the same
psychological state has diverse realizations. But we can always disagree with the
functional taxonomy, and claim there are psychological differences at another
level of description. (2004, p. 203)

Thus he relates that while the two types of eyes have similar structure in certain respects,
both consisting of a spherical shell, lens and retina, they use different kinds of visual
pigments in their photoreceptors, as well as having different numbers of them, the
octopus having one in contrast to the human eye which has four. They also have different
retinas. The human retina, with its rods and cones, allows the eye to focus light by
bending the lens, thereby changing its shape. The octopus eye, with rhabdomeres instead
of rods and cones, focuses light by moving the lens backwards and forwards within the
177
shell. All these factors show up as differences in output, not just structure. The octopus,
having only a single pigment, is colorblind, while its receptor’s unique structure allows it
to perceive the plane of polarized light. Retinal differences likewise make for functional
differences, with very little information processing occurring on the octopus’s retina,
unlike the case of the human retina. This produces differences in stimuli and reaction
times. So the two eyes might be similar, but when described with a suitably fine grain, he
contends, they come out type distinct. In the result they are both physically and
cognitively diverse, and so not genuine examples of MR.

Notice again that, contrary to what is claimed, it has not been demonstrated that
type-type identity prevails here after all (on the understanding that the kind camera
eyehuman reduces to its distinct neural type, and the kind camera eyemollusc in turn reduces
to its distinct neural type). If anything what this foray into mollusc visual physiology
succeeds in showing is that, relative to the kind camera eye, human camera eyes and
octopus camera eyes count as distinct realizations(!), for, assuming Shapiro’s causal
relevance criterion applies, human camera eyes achieve the function of camera vision
differently to the way octopus camera eyes achieve this function. Were we to attend to the
original inquiry, which concerned whether human eyes and octopus eyes count as distinct
realizations of the kind eye, Shapiro’s own response, for what it’s worth, is clear (2000,
pp. 645-646): here we do seem to confront a genuine case of type-type identity, as
Putnam himself assumed, because, relative to the function of vision (not camera vision),
both humans and molluscs achieve the function the same way (namely, by camera
vision!). Differences that would be relevant at the neural level between humans and
molluscs when asking how camera vision is achieved can be conveniently screened off
when the question is how vision, as distinct from camera vision, is achieved. Again if pain
or hunger were the kind in question, it seems more likely than not that we would confront
a case of MR (unlike with vision), as we conjectured earlier. Explanatory context dictates
the function of interest, and the function is one that we have to assume is common to the
tokens in question in order to get the inquiry into MR off the ground. Indeed if Shapiro’s
analysis is correct, with MR we’re always asking how some common function is achieved
by different tokens that do that thing. Where there is no common function the question of
MR cannot so much as arise. The fact that the question does arise in all the cases we’ve
considered is a powerful indication that we’re dealing with functions which all the

178
relevant tokens actually share. The grains strategy confuses matters by suggesting that in
many cases involving putative MR kinds, psychological states can be individuated using a
finer grain of description. But if what I have been saying is right, this is not the proper
way to refute a putative case of MR.

That mine is the correct assessment of the situation is not only attested to by
Shapiro’s analysis of MR, but also by the fact that it avoids the very mug’s game Polger
sought to eschew by embracing the grains strategy in the first place. If for any putative
MR kind I am free to cavil with the choice of your size of grain (“oh, that’s far too coarse
for psychology,” or “now that’s really not coarse enough for neuroscience”), how is the
resulting game any less of a mug’s game than the one we were trapped in at the start? I
myself have played a few of these games with philosophers. No one wins. Couch’s
remarks are telling: “we can always disagree with the functional taxonomy, and claim
there are psychological differences at another level of description.” So the game goes on.4
Yes, it is true that Bechtel and Mundale don’t overtly admit to thinking their criterion
eradicates MR kinds completely; and yet it would have been interesting to see them
pointing out even a single instance of an MR kind they think rightly counts as such. But
alas they don’t.5 All their examples come out as type-type identities after being subjected
to their test—even where it is highly likely that Shapiro’s more precise test would lead us
to conclude otherwise.

9.4 SUMMARY

There’s a real problem with the grain requirement. The central difficulty is that in the
terms in which it’s been put it is largely unworkable, and at best no more than a loose
metaphor. For a recommendation intended to serve at least in part as a methodological
reform, this is clearly unsatisfactory. I don’t deny that Bechtel and Mundale were onto
something. But whatever value their insight into MR might have has been obscured by
their unfortunate formulation of the issue. Moreover, as I have tried to show, the

4In subsequent work, Couch (2009a; 2009b) has been more careful in his remarks, and has brought out
explicitly the importance of scientific taxonomy in the individuation of both neural and psychological kinds.
Still I notice that in these papers he doesn’t rely on granularity arguments, and indeed is even mildly critical
of them (2009a, p. 267). When grains hold sway, MR stands little chance of receiving a fair hearing.
5By contrast, Couch (2009b, p. 514) puts forward human eyes and pigeon eyes as plausible candidates for
MR, and comments: “Accepting that this example is plausible is a revision from an earlier view of mine.”

179
formulation is unfortunate not just because it happens to be unworkable. More
worryingly, the argument from grains distorts the truth about MR by encouraging the
view that mind-brain identity comes for free once we invoke the “same grain” of
description across both realized and realizing kinds. But when the insight to which this
locution seems to point is expressed in terms that are intelligible and empirically tractable
(namely, Shapiro’s causal relevance criterion), mind-brain identity seems anything but a
fait accompli. Grains talk makes it tempting to think MR is easier to refute than it in fact
is. It is certainly true, as Bechtel and Mundale acknowledge, that context fixes the choice
of grain (where by “grain” we mean the respect under which we seek to compare a set of
tokens); but we are not ipso facto obliged to employ a consistent grain across realized and
realizing kinds (since this is just about meaningless as far as a researcher into these
matters would be concerned and raises a host of difficulties beside). Rather than matching
grains, what MR really behooves us to do is to apply a principled method for adjudicating
upon differences between tokens of a functional kind. Shapiro’s work on MR shows us
how to approach this important task.

180
 10
_____

Conclusion

This thesis has been concerned with a specific feature of the organization of biological
systems. Livers, hair, eyes, skin, hearts—each exhibits in its own way a certain richness of
inner structure which it would be foolish to suppose stops the moment one reaches the
brain. Happily, no one denies the brain’s intricacy of structure and function. The debate
has always been over what form this complexity takes. The most influential answer to this
question over the past 60 years—and the most controversial for at least the past 30—is
that the mind is composed of modules. I took the canonical expression of this concept
from Fodor, but isolated it from some of its peculiarities, most especially the notion of
strict domain specificity and sensory transduction. I put this refined concept to the test
and ended up with a mixed bag of results. Fodorian modules survive in some ways, but
die in others. The modules that survive are functionally and anatomically exiguous when
set against those postulated by mainstream evolutionary psychologists. They do not
handle gross cognitive functions. In effect they are the columns which Vernon
Mountcastle originally hypothesized some 60 years ago, and form part of the well-known
“columnar hypothesis” in neuroscience. These modules extend throughout the cortex, so
there can be no real sense in which central systems are not modular. This is to say that the
cortex appears to be modular in the general sense that it exhibits a limited (and as yet
undetermined) degree of functional specialization consistent with the reuse of neural
resources. There seems to be no particular difference in this regard between peripheral
systems and central systems. Low-level sensory systems appear as reliant on domain-
general mechanisms as central ones. Perceptual and linguistic systems do not exhibit the
defining characteristics of Fodorian modularity.

181
 Still, the fate of this revised notion of modularity is not certain. The main issue
confronting modularity in this revised sense is the effect of neural network context on
local function. At some point the effects of context are so strong that the degree of
specialization required for modularity is not able to be met. This does not mean that such
brain regions are infinitely plastic, prey entirely to the whims of the neural network in
which they find themselves: their plasticity is actually impressively constrained, and they
exhibit a considerable degree of developmental robustness. Nevertheless the extent of
strong context effects may turn out to be great enough to put a decisive end to
modularity’s long reign. Recent work in neurobiology is thus forcing a redefinition of the
architecture of cognition, principally in terms of patterns of interconnectivity, partial
specialization and emergent specialization. As Giordana Grossi summarizes recent trends:

…cognitive and brain systems that are specialized in adults develop in a highly
interconnected brain where regions co-develop with other brain regions, not in
isolation. What a brain region or neuron does, in terms of function, depends on its
interaction with other regions and neurons[;] it even depends on the state of
distributed neural networks….Within this framework, the specialization of neural
systems (modularity) assumes a different meaning, one that is anchored into the
physical system of a developing organism…. (2014, p. 346)

Turning to the language module, we saw that there probably is no such thing, not
at any rate in the conventional sense, and that dissociations which are otherwise
compelling evidence of domain specificity can be adequately explained by the
Redundancy Model, which predicates functional inheritance across tasks and task
categories even when the tasks are implemented in spatially segregated neural networks.

All up, this is a brave new world. It offers a clearer, cleaner and far more realistic
picture of how the mind works. It is respectful of advances in psychology and philosophy
over the past half century, but is anchored firmly in the neurobiological evidence. It
strikes what I think is an ideal balance between different approaches to the investigation
of the mind/brain. I commend it unreservedly.

182
Bibliography

Ackerman, J.M., Nocera, C.C. & Bargh, J.A. (2010) Incidental haptic sensations
influence social judgments and decisions. Science 328: 1712-1715.

Aizawa, K. (2009) Neuroscience and multiple realization: A reply to Bechtel and

Mundale. Synthese 167(3): 493-510.

Aizawa, K. & Gillett, C. (2009) Levels, individual variation, and massive multiple
realization in neurobiology. In: The Oxford Handbook of Philosophy and Neuroscience,
ed. J. Bickle, pp. 539-81. New York: Oxford University Press.

Altun Z.F. & Hall D.H. (2011) Nervous system, general description. In: WormAtlas, eds.
Z.F. Altun, L.A. Herndon, C. Crocker, R. Lints & D.H. Hall. <http://www. wormatlas.
org/ver1/handbook/contents.htm>

Amaral, D.G. & Strick, P.L. (2013) The organization of the central nervous system. In:
Principles of neural science, eds. E.R. Kandel, J.H. Schwartz, T.M. Jessell, S.A.
Siegelbaum & A.J. Hudspeth, pp. 337-355. New York: McGraw-Hill.

Anderson, M.L. (2007a) Evolution of cognitive function via redeployment of brain areas.
The Neuroscientist 13: 13-21.

⎯⎯(2007b) Massive redeployment, exaptation, and the functional integration of

cognitive operations. Synthese 159(3): 329-345.

⎯⎯(2007c) The massive redeployment hypothesis and the functional topography of the
brain. Philosophical Psychology 21(2): 143-174.

⎯⎯(2008) Circuit sharing and the implementation of intelligent systems. Connection

Science 20(4): 239-251.

⎯⎯(2010) Neural reuse: A fundamental organizational principle of the brain. Behavioral

and Brain Sciences 33(4): 245-266; discussion 266-313.

⎯⎯(2014) After phrenology: Neural reuse and the interactive brain. Cambridge, MA:
MIT Press.

⎯⎯(2015) Mining the brain for a new taxonomy of the mind. Philosophy Compass
10(1): 68-77.

⎯⎯(2016) Précis of After phrenology: Neural reuse and the interactive brain. Behavioral
and Brain Sciences 39: 1-10; discussion 10-45.

Anderson, M.L. & Finlay, B.L. (2014) Allocating structure to function: The strong links
between neuroplasticity and natural selection. Frontiers in Human Neuroscience 7: 1-16.

183
Apperly, I.A., Samson, D., Carroll, N., Hussain, S. & Humphreys, G. (2006) Intact first-
and second-order false belief reasoning in a patient with severely impaired grammar.
Social Neuroscience 1(3-4): 334-348.

Ariew, A. (1996) Innateness and canalization. Philosophy of Science 63: S19-S27.

⎯⎯(1999) Innateness is canalization: In defense of a developmental account of

innateness. In: Where biology meets psychology: Philosophical essays, ed. V.G.
Hardcastle, pp. 117-138. Cambridge, MA: MIT Press.

⎯⎯(2007) Innateness. In: Handbook of the philosophy of science: Philosophy of

Biology, eds. M. Matthen & C. Stephens, pp. 567-584. Dordrecht: Elsevier.

Avital, E. & Jablonka, E. (2000) Animal Traditions: Behavioural Inheritance in Evolution.

Cambridge: Cambridge University Press.

Bach-y-Rita, P. (2004) Emerging concepts of brain function. Journal of Integrative

Neuroscience 4(2): 183-205.

Badcock, P.B., Ploeger, A. & Allen, N.B. (2016) After phrenology: Time for a paradigm
shift in cognitive science. Behavioral and Brain Sciences 39: 10-11.

Bargh, J.A., Williams, L.E., Huang, J.Y., Song, H. & Ackerman, J.M. (2010) From the
physical to the psychological: Mundane experiences influence social judgment and
interpersonal behavior. Behavioral and Brain Sciences 33(4): 267-268.

Barrett, H.C. (2006) Modularity and design reincarnation. In: The Innate Mind Volume
2: Culture and Cognition, eds. P. Carruthers, S. Laurence & S.P. Stich, pp. 199-217.
New York: Oxford University Press.

Barrett, H.C. & Kurzban, R. (2006) Modularity in cognition: Framing the debate.
Psychological Review 113(3): 628-647.

Bates, E. (1999) Plasticity, localization, and language development. In: The changing
nervous system: Neurobehavioral consequences of early brain disorders, eds. S.H.
Broman & J.M. Fletcher, pp. 214-253. New York: Oxford University Press.

Bateson, P. & Mameli, M. (2007) The innate and the acquired: Useful clusters or a
residual distinction from folk biology? Developmental Psychobiology 49: 818-831.

Bechtel, W. (2008a) Mechanisms in cognitive psychology: What are the operations?

Philosophy of Science 75(5): 983-994.

⎯⎯(2008b) Mental Mechanisms: Philosophical Perspectives on Cognitive

Neuroscience. London: Routledge.

Bechtel, W. & Mundale, J. (1999) Multiple realizability revisited: Linking cognitive and
neural states. Philosophy of Science 66(2): 175-207.

184
Bergeron, V. (2007) Anatomical and functional modularity in cognitive science: Shifting
the focus. Philosophical Psychology 20(2): 175-195.

⎯⎯(2008) Cognitive architecture and the brain: Beyond domain-specific functional

specification. Unpublished doctoral dissertation, Department of Philosophy, University
of British Columbia. Available at: http://circle.ubc.ca/handle/2429/2711.

Berwick, R.C. & Chomsky, N. (2016) Why only us: Language and evolution. Cambridge,
MA: MIT Press.

Bickle, J. (1998) Psychoneural reduction: The new wave. Cambridge, MA: MIT Press.

⎯⎯(2003) Philosophy and neuroscience: A ruthlessly reductive account. Dordrecht:

Springer.

⎯⎯(2010) Has the last decade of challenges to the multiple realization argument
provided aid and comfort to psychoneural reductionists? Synthese 177(2): 247-260.

⎯⎯ed. (2009) The Oxford handbook of philosophy and neuroscience. New York:
Oxford University Press.

Binkofski, F., Amunts, K., Stephan, K.M., Posse, S., Schormann, T., Freund, H.-J.,
Zilles, K. & Seitz, R.J. (2000) Broca’s region subserves imagery of motion: A combined
cytoarchitectonic and fMRI study. Human Brain Mapping 11: 273-285.

Bloom, P. (2000) How children learn the meanings of words. Cambridge, MA: MIT
Press.

Brattico, P. & Liikkanen, L. (2009) Rethinking the Cartesian theory of linguistic

productivity. Philosophical Psychology 22(3): 251-279.

Boone, W. & Piccinini, G. (2016) The cognitive neuroscience revolution. Synthese 193:
1509-1534.

Boyd, R., Richerson, P.J. & Henrich, J. (2011) The cultural niche: Why social learning is
essential for human adaptation. Proceedings of the National Academy of Sciences of the
United States of America 108(Supplement 2): 10918-10925.

Bregman, A.S. & Pinker, S. (1978) Auditory streaming and the building of timbre.
Canadian Journal of Psychology 32(1): 19-31.

Bressler, S.L. (1995) Large-scale cortical networks and cognition. Brain Research
Reviews 20: 288-304.

Bridgeman, B. (2010) Neural reuse implies distributed coding. Behavioral and Brain
Sciences 33(4): 269-270.

Brighton, H., Kirby, S. & Smith, K. (2005) Cultural selection for learnability: Three
principles underlying the view that language adapts to be learnable. In: Language origins:

185
Perspectives on evolution, ed. M. Tallerman, pp. 291-309. Oxford: Oxford University
Press.

Bullmore, E. & Sporns, O. (2012) The economy of brain network organization. Nature
Reviews Neuroscience 13(5): 336-349.

Buonomano, D.V. & Merzenich, M.M. (1998) Cortical plasticity: From synapses to
maps. Annual Review of Neuroscience 21: 149-186.

Burnston, D.C. (2016) A contextualist approach to functional localization in the brain.

Biology and Philosophy 31: 527-550.

Buxhoeveden, D.P. & Casanova, M.F. (2002) The minicolumn hypothesis in

neuroscience. Brain 125: 935-951.

Caldarelli, G. & Catanzaro, M. (2012) Networks: A very short introduction. Oxford:

Oxford University Press.

Carandini, M. (2015) From circuits to behavior: A bridge too far? In: The future of the
brain, eds. G. Marcus & J. Freeman, pp. 177-185. Princeton: Princeton University Press.

Carruthers, P. (2006) The architecture of the mind: Massive modularity and the
flexibility of thought. Oxford: Oxford University Press.

⎯⎯(2008) Précis of The architecture of the mind: Massive modularity and the flexibility
of thought. Mind and Language 23(3): 257-262.

Casasanto, D. & Dijkstra, K. (2010) Motor action and emotional memory. Cognition
115(1): 179-185.

Chao, L.L. & Martin, A. (2000) Representation of manipulable man-made objects in the
dorsal stream. NeuroImage 12: 478-484.

Chomsky, N. (1956) Three models for the description of language. IRE Transactions on
Information Theory IT-2: 113-124. Reprinted, with corrections, in: Readings in
mathematical psychology, Vol. II, eds. R.D. Luce, R.R. Bush & E. Gallanter, pp. 105-
124. New York: John Wiley & Sons, 1965. References are to the reprint.

⎯⎯(1957) Syntactic structures. Berlin: Mouton de Gruyter.

⎯⎯(1965) Aspects of the theory of syntax. Cambridge, MA: MIT Press.

⎯⎯(1975) Reflections on language. New York: Pantheon.

⎯⎯(1979) Language and responsibility (with Mitsou Ronat). New York: Pantheon.

⎯⎯(1980a) Rules and representations. New York: Columbia University Press.

186
⎯⎯(1980b) Rules and representations. Behavioral and Brain Sciences 3(1): 1-15;
discussion 15-61.

⎯⎯(1988) Language and problems of knowledge: The Managua lectures. Cambridge,

MA: MIT Press.

⎯⎯(1995) The minimalist program. Cambridge, MA: MIT Press.

⎯⎯(2002) On nature and language. New York: Cambridge University Press.

⎯⎯(2005) Three factors in language design. Linguistic Inquiry 36(1): 1-22.

⎯⎯(2006) Language and mind. 3rd ed. New York: Cambridge University Press.

⎯⎯(2010) Some simple evo devo theses: How true might they be for language? In: The
evolution of human language: Biolinguistic perspectives, eds. R.K. Larson, V. Déprez &
H. Yamakido, pp. 45-72. New York: Cambridge University Press.

⎯⎯(2016) Minimal computation and the architecture of language. Chinese Semiotic

Studies 12(1): 13-24.

Christiansen, M.H. & Chater, N. (2016) Creating language: Integrating evolution,

acquisition, and processing. Cambridge, MA: MIT Press.

Clark, A. (2013) Whatever next? Predictive brains, situated agents, and the future of
cognitive science. Behavioral and Brain Sciences 36(3): 181-204; discussion 204-253.

Clark, E. (2009) First language acquisition. 2nd ed. Cambridge: Cambridge University
Press.

Coase, R.H. (1937) The nature of the firm. Economica New Series 4(16): 386-405.

Cohen, L.G, Celnik, P., Pascual-Leone, A., Corwell, B., Faiz, L., Dambrosia, J., Honda,
M., Sadato, N., Gerloff, C., Catalá, M.D. & Hallett, M. (1997) Functional relevance of
cross-modal plasticity in blind humans. Nature 389(6647): 180-183.

Cole, M.W., Reynolds, J.R., Power, J.D., Repovs, G., Anticevic, A. & Braver, T.S.
(2013) Multi-task connectivity reveals flexible hubs for adaptive task control. Nature
Neuroscience 16(9): 1348-1355.

Collins, J. (2004) Faculty disputes. Mind and Language 19(5): 503-533.

⎯⎯(2005) Nativism: In defense of a biological understanding. Philosophical Psychology

18(2): 157-177.

⎯⎯(2008) Chomsky: A guide for the perplexed. London: Continuum.

Coltheart, M. (1999) Modularity and cognition. Trends in Cognitive Sciences 3(3): 115-
120.

187
⎯⎯(2004) Brain imaging, connectionism, and cognitive neuropsychology. Cognitive
Neuropsychology 21(1): 21-25.

⎯⎯(2011) Methods for modular modelling: Additive factors and cognitive

neuropsychology. Cognitive Neuropsychology 28(3-4): 224-240.

Cosmides, L. & Tooby, J. (1994) Origins of domain specificity: The evolution of

functional organization. In: Mapping the world: Domain specificity in cognition and
culture, eds. L. Hirschfield & S. Gelman, pp. 85-116. New York: Cambridge University
Press.

Couch, M.B. (2004) A defense of Bechtel and Mundale. Philosophy of Science 71(2):
198-204.

⎯⎯(2009a) Functional explanation in context. Philosophy of Science 76(2): 253-269.

⎯⎯(2009b) Multiple realization in comparative perspective. Biology and Philosophy 24:

505-519.

Cowie, F. (2008) Innateness and language. In: The Stanford Encyclopedia of Philosophy,
winter 2016, ed. E.N. Zalta. <http://plato.stanford.edu/archives/win2016/entries/
innateness-language/>

Craver, C.F. (2007) Explaining the brain. Oxford: Oxford University Press.

Croft, W. & Cruise, D.A. (2004) Cognitive linguistics. Cambridge: Cambridge University
Press.

da Costa, N.M. & Martin, K.A.C. (2010) Whose cortical column would that be? Frontiers
in Neuroanatomy 4(5): 1-10.

Damasio, A.R. & Tranel, D. (1993) Nouns and verbs are retrieved with differently
distributed neural systems. Proceedings of the National Academy of Sciences of the
United States of America 90: 4957-4960.

Damasio, H., Grabowski, T.J., Tranel, D., Hichwa, R.D. & Damasio, A.R. (1996) A
neural basis for lexical retrieval. Science 380: 499-505.

Danelli, L., Cossu, G., Berlingeri, M., Bottini, G., Sberna, M. & Paulesu, E. (2013) Is a
lone right hemisphere enough? Neurolinguistic architecture in a case with a very early left
hemispherectomy. Neurocase 19(3): 209-231.

de Boer, B. (2016) Evolution of speech and evolution of language. Psychonomic Bulletin

and Review 2016: 1-5 doi: 10.3758/s13423-016-1130-6.

Deacon, T.W. (1997) The symbolic species: The co-evolution of language and the brain.
New York: Norton.

188
Decety, J., Grèzes, J., Costes, N., Perani, D., Jeannerod, M., Procyk, E., Grassi, F. &
Fazio, F. (1997) Brain activity during observation of actions. Influence of action content
and subject’s strategy. Brain 120(10): 1763-1777.

Dehaene, S. (2005) Evolution of human cortical circuits for reading and arithmetic: The
“neuronal recycling” hypothesis. In: From monkey brain to human brain, eds. S.
Dehaene, J.R. Duhamel, M.D. Hauser & G. Rizzolatti, pp. 133-157. Cambridge, MA:
MIT Press.

Dehaene, S., Bossini, S. & Giraux, P. (1993) The mental representation of parity and
numerical magnitude. Journal of Experimental Psychology: General 122: 371-396.

Dennett, D.C. (1991) Consciousness explained. London: Penguin.

Doidge, N. (2007) The brain that changes itself. New York: Penguin.

Dor, D. & Jablonka, E. (2010) Canalization and plasticity in the evolution of linguistic
communication: An evolutionary developmental approach. In: The evolution of human
language: Biolinguistic perspectives, eds. R.K. Larson, V. Déprez & H. Yamakido, pp.
135-147. New York: Cambridge University Press.

Dunn, M., Greenhill, S.J., Levinson, S.C. & Gray, R.D. (2011) Evolved structure of
language shows lineage-specific trends in word-order universals. Nature 473: 79-82.

Elbert, T., Pantev, C., Wienbruch, C., Rockstroh, B. & Taub, E. (1995) Increased
cortical representation of the fingers of the left hand in string players. Science 270: 305-
307.

Eliasmith, C. (2015) Building a behaving brain. In: The future of the brain, eds. G.
Marcus & J. Freeman, pp. 125-136. Princeton: Princeton University Press.

Evans, N. & Levinson, S.C. (2009) The myth of language universals: Language diversity
and its importance for cognitive science. Behavioral and Brain Sciences 32(5): 429-448;
discussion 448-492.

Evans, V. & Green, M. (2006) Cognitive linguistics: An introduction. Edinburgh:

Edinburgh University Press.

Everett, D. (2012) Language: The cultural tool. New York: Vintage.

Fedorenko, E., Behr, M.K. & Kanwisher, N. (2011) Functional specificity for high-level
linguistic processing in the human brain. Proceedings of the National Academy of
Sciences of the United States of America 108(39): 16428-16433.

Fedorenko, E., Duncan, J. & Kanwisher, N. (2012) Language-selective and domain-

general regions lie side by side within Broca’s area. Current Biology 22(21): 2059-2062.

Fedorenko, E. & Thompson-Schill, S.L. (2014) Reworking the language network.

Trends in Cognitive Sciences 18(3): 120-126.

189
Fisher, S.E. (2015) Translating the genome in human neuroscience. In: The future of the
brain, eds. G. Marcus & J. Freeman, pp. 149-158. Princeton: Princeton University Press.

Fisher, S.E. & Scharff, C. (2009) FOXP2 as a molecular window into speech and
language. Trends in Genetics 25: 166-177.

Fitch, W.T. (2010) Three meanings of “recursion”: Key distinctions for biolinguistics.
In: The evolution of human language: Biolinguistic perspectives, eds. R.K. Larson, V.
Déprez & H. Yamakido, pp. 73-90. New York: Cambridge University Press.

Fitch, W.T., Hauser, M.D. & Chomsky, N. (2005) The evolution of the language faculty:
Clarifications and implications. Cognition 97: 179-210.

Fodor, J.A. (1975) The language of thought. Cambridge, MA: Harvard University Press.

⎯⎯(1983) The modularity of mind: An essay on faculty psychology. Cambridge, MA:

MIT Press.

Fodor, J.A., Bever, T. & Garrett, M. (1974) The psychology of language: An

introduction to psycholinguistics and generative grammar. New York: McGraw-Hill.

Frost, R., Armstrong, B.C., Siegelman, N. & Christiansen, M.H. (2015) Domain
generality versus modality specificity: The paradox of statistical learning. Trends in
Cognitive Sciences 19(3): 117-125.

Fuchs, E. & Flügge, G. (2014) Adult neuroplasticity: More than 40 years of research.
Neural Plasticity 2014: 1-10 doi: 10.1155/2014/541870.

Gall, F.J. & Spurzheim, J.C. (1835) On the function of the brain and each of its parts.
Boston: Marsh Capen and Lyon.

Gauthier, I., Curran, T., Curby, K.M. & Collins, D. (2003) Perceptual interference
supports a non-modular account of face processing. Nature Neuroscience 6(4): 428-432.

Gauthier, I., Skudlarski, P. Gore, J.C. & Anderson, A.W. (2000) Expertise for cars and
birds recruits brain areas involved in face recognition. Nature Neuroscience 3(2): 191-
197.

Gazzaniga, M.S. (1989) Organization of the human brain. Science 245(4921): 947-952.

Gerrans, P. (2014) The measure of madness: Philosophy of mind, cognitive neuroscience,

and delusional thought. Cambridge, MA: MIT Press.

Gilbert, C.D. (2013) The constructive nature of visual processing. In: Principles of neural
science, eds. E.R. Kandel, J.H. Schwartz, T.M. Jessell, S.A. Siegelbaum & A.J.
Hudspeth, pp. 556-576. New York: McGraw-Hill.

Gillett, C. (2002) The dimensions of realization: A critique of the standard view. Analysis
62(4): 316-323.

190
⎯⎯(2003) The metaphysics of realization, multiple realizability, and the special
sciences. Journal of Philosophy 100(11): 591-603.

Glenberg, A.M., Brown, M. & Levin, J.R. (2007) Enhancing comprehension in small
reading groups using a manipulation strategy. Contemporary Educational Psychology 32:
389-399.

Glenberg, A.M. & Kaschak, M.P. (2002) Grounding language in action. Psychonomic
Bulletin and Review 9: 558-565.

Glenberg, A.M., Sato, M. & Cattaneo, L. (2008) Use-induced motor plasticity affects the
processing of abstract and concrete language. Current Biology 18(7): R290-291.

Glickstein, M. (2014) Neuroscience: A historical introduction. Cambridge, MA: MIT

Press.

Gold, I. & Roskies, A.L. (2008) Philosophy of neuroscience. In: The Oxford Handbook
of Philosophy of Biology, ed. M. Ruse, pp. 349-380. New York: Oxford University Press.

Goldberg, A.E. (2003) Constructions: A new theoretical approach to language. Trends in

Cognitive Sciences 7(5): 219-224.

Graziano, M.S.A., Taylor, C.S.R., Moore, T. & Cooke, D.F. (2002) The cortical control
of movement revisited. Neuron 36: 349-362.

Greenfield, P.M. (1991) Language, tools and brain: The ontogeny and phylogeny of
hierarchically organized sequential behavior. Behavioral and Brain Sciences 14(4): 531-
551; discussion 551-595.

Greenhill, S.J., Atkinson, Q.D., Meade, A. & Gray, R.D. (2010) The shape and tempo of
language evolution. Proceedings of the Royal Society B: Biological Sciences 277: 2443-
2450.

Griffiths, P.E. (2002) What is innateness? Monist 85: 70-85.

Griffiths, P.E. & Machery, E. (2008) Innateness, canalization, and “biologicizing the
mind.” Philosophical Psychology 21(3): 397-414.

Griffiths, P.E., Machery, E. & Linquist, S. (2009) The vernacular concept of innateness.
Mind and Language 24(5): 605-630.

Grossi, G. (2014) A module is a module is a module: Evolution of modularity in

evolutionary psychology. Dialectical Anthropology 38: 333-351.

Gruber, O. (2002) The co-evolution of language and working memory capacity in the
human brain. In: Mirror neurons and the evolution of brain and language, eds. M.I.
Stamenov & V. Gallese, pp. 77-86. Amsterdam: John Benjamins.

Guida, A., Campitelli, G. & Gobet, F. (2016) Becoming an expert: Ontogeny of expertise
as an example of neural reuse. Behavioral and Brain Sciences 39: 13-15.
191
Hagoort, P. & Indefrey, P. (2014) The neurobiology of language beyond single words.
Annual Review of Neuroscience 37: 347-362.

Hanson, S.J., Matsuka, T. & Haxby, J.V. (2004) Combinatorial codes in ventral temporal
lobe for object recognition: Is there a “face” area? NeuroImage 23(1): 156-166.

Hardcastle, V.G. & Stewart, C.M. (2009) fMRI: A modern cerebrascope? The case of
pain. In: The Oxford handbook of philosophy and neuroscience, ed. J. Bickle, pp. 179-
199. New York: Oxford University Press.

Hauser, M.D., Chomsky, N. & Fitch, W.T. (2002) The faculty of language: What is it,
who has it and how did it evolve? Science 298(5598): 1569-1579.

Hawrylycz, M. (2015) Building atlases of the brain. In: The future of the brain, eds. G.
Marcus & J. Freeman, pp. 3-16. Princeton: Princeton University Press.

Haxby, J.V., Gobbini, M.I., Furey, M.L., Ishai, A., Schouten, J.L. & Pietrini, P. (2001)
Distributed and overlapping representations of faces and objects in ventral temporal
cortex. Science 293: 2425-2430.

Hebb, D.O. (1949) The organization of behavior: A neuropsychological theory. New

York: Wiley.

Hickok, G. & Poeppel, D. (2000) Towards a functional neuroanatomy of speech

perception. Trends in Cognitive Sciences 4(4): 131-138.

Hirst, P., Khomami, P.J., Gharat, A. & Zangenehpour, S. (2012) Cross-modal

recruitment of primary visual cortex by auditory stimuli in the nonhuman primate brain:
A molecular mapping study. Neural Plasticity 2012: 1-11 doi: 10.1155/2012/197264.

Horst, S. W. (2011) Laws, mind, and free will. Cambridge, MA: MIT Press.

Hsu, H.J., Tomblin, J.B. & Christiansen, M.H. (2014) Impaired statistical learning of
non-adjacent dependencies in adolescents with specific language impairment. Frontiers in
Psychology 5: 175.

Hubbard, E.M., Piazza, M., Pinel, P. & Dehaene, S. (2005) Interactions between number
and space in parietal cortex. Nature Reviews Neuroscience 6(6): 435-448.

Jackendoff, R. (2007) A Parallel Architecture perspective on language processing. Brain

Research 1146: 2-22.

Jacobs, J.A. (1999) Computational studies of the development of functionally specialized

neural modules. Trends in Cognitive Sciences 3(1): 31-38.

Jilk, D.J., Lebiere, C., O’Reilly, R.C. & Anderson, J.R. (2008) SAL: An explicitly
pluralistic cognitive architecture. Journal of Experimental and Theoretical Artificial
Intelligence 20: 197-218.

192
Jungé, J.A. & Dennett, D.C. (2010) Multi-use and constraints from original use.
Behavioral and Brain Sciences 33(4): 277-278.

Jusczyk, P.W. & Cohen, A. (1985) What constitutes a module? Behavioral and Brain
Sciences 8(1): 20-21.

Kaan, E. & Stowe, L.A. (2002) Storage and computation in the brain: A neuroimaging
perspective. In: Storage and computation in the language faculty, eds. S.G. Nooteboom,
F. Weerman & F.N.K. Wijnen, pp. 257-298. Dordrecht: Kluwer.

Kaan, E. & Swaab, T.Y. (2002) The brain circuitry of syntactic comprehension. Trends
in Cognitive Sciences 6(8): 350-356.

Kandel, E.R., Barres, B.A. & Hudspeth A.J. (2013) Nerve cells, neural circuitry, and
behavior. In: Principles of neural science, eds. E.R. Kandel, J.H. Schwartz, T.M. Jessell,
S.A. Siegelbaum & A.J. Hudspeth, pp. 21-38. New York: McGraw-Hill.

Kandel, E.R. & Hudspeth, A.J. (2013) The brain and behavior. In: Principles of neural
science, eds. E.R. Kandel, J.H. Schwartz, T.M. Jessell, S.A. Siegelbaum & A.J.
Hudspeth, pp. 5-20. New York: McGraw-Hill.

Kandel, E.R. & Siegelbaum, S.A. (2013) Cellular mechanisms of implicit memory storage
and the biological basis of individuality. In: Principles of neural science, eds. E.R.
Kandel, J.H. Schwartz, T.M. Jessell, S.A. Siegelbaum & A.J. Hudspeth, pp. 1461-1486.
New York: McGraw-Hill.

Kanwisher, N., McDermott, J. & Chun, M. (1997) The fusiform face area: A module in
human extrastriate cortex specialized for face perception. Journal of Neuroscience 17(11):
4302-4311.

Karmiloff-Smith, A. (1992) Beyond modularity: A developmental perspective on

cognitive science. Cambridge, MA: MIT Press.

⎯⎯(1994) Précis of Beyond modularity: A developmental perspective on cognitive

science. Behavioral and Brain Sciences 17(4): 693-706; discussion 707-745.

Kim, J. (1992) Multiple realization and the metaphysics of reduction. Philosophy and
Phenomenological Research 52(1): 1-26.

Kim, S. (2002) Testing multiple realizability: A discussion of Bechtel and Mundale.

Philosophy of Science 69(4): 606–610.

Kirby, S., Dowman, M. & Griffiths, T.L. (2007) Innateness and culture in the evolution
of language. Proceedings of the National Academy of Sciences of the United States of
America 104: 5241-5245.

Klein, C. (2008) An ideal solution to disputes about multiply realized kinds. Philosophical
Studies 140(2): 161-177.

193
⎯⎯(2010) Redeployed functions versus spreading activation: A potential confound.
Behavioral and Brain Sciences 33(4): 280-281.

⎯⎯(2012) Cognitive ontology and region- versus network-oriented analyses. Philosophy

of Science 79(5): 952-960.

Koch, C. (2015) Project MindScope. In: The future of the brain, eds. G. Marcus & J.
Freeman, pp. 25-39. Princeton: Princeton University Press.

Kolb, B., Gibb. R. & Gonzalez, C.L.R. (2001) Cortical injury and neuroplasticity during
brain development. In: Toward a theory of neuroplasticity, eds. C.A. Shaw & J.C.
McEachern, pp. 223-243. Philadelphia: Psychology Press.

Krubitzer, L. (1995) The organization of neocortex in mammals: Are species differences

really so different? Trends in Neurosciences 18(9): 408-417.

Kuhl, P.K. & Damasio, A.R. (2013) Language. In: Principles of neural science, eds. E.R.
Kandel, J.H. Schwartz, T.M. Jessell, S.A. Siegelbaum & A.J. Hudspeth, pp. 1353-1372.
New York: McGraw-Hill.

Lai, C.S.L., Fisher, S.E., Hurst, J.A., Vargha-Khadem, F. & Monaco, A.P. (2001) A
forkhead-domain gene is mutated in a severe speech and language disorder. Nature 413:
519-523.

Laland, K.N. (2016) The origins of language in teaching. Psychonomic Bulletin and
Review 2016: 1-7 doi: 10.3758/s13423-016-1077-7.

Laland, K.N. Sterelny, K., Odling-Smee, J., Hoppit, W. & Uller, T. (2011) Cause and
effect in biology revisited: Is Mayr’s proximate-ultimate dichotomy still useful? Science
334(6062): 1512-1516.

Laland, K.N, Uller, T., Feldman, M.W., Sterelny, K., Müller, G.B., Moczek, A.,
Jablonka, E. & Odling-Smee, J. (2015) The extended evolutionary synthesis: Its
structure, assumptions and predictions. Proceedings of the Royal Society B 282:
20151019.

Landon, B. (2013) Building great sentences: How to write the kinds of sentences you love
to read. Penguin: New York.

Laurence, S. & Margolis, E. (2015) Concept nativism and neural plasticity. In: The
conceptual mind: New directions in the study of concepts, eds. E. Margolis & S.
Laurence, pp. 117-147. Cambridge, MIT: MIT Press.

Leiber, J. (2006) Turing’s golden: How well Turing’s work stands today. Philosophical
Psychology 19(1): 13-46.

Leise E.M. (1990) Modular construction of nervous systems: A basic principle of design
for invertebrates and vertebrates. Brain Research Reviews 15: 1-23.

194
Leo, A., Bernardi, G., Handjaras, G., Bonino, D., Ricciardi, E. & Pietrini, P. (2012)
Increased BOLD variability in the parietal cortex and enhanced parieto-occipital
connectivity during tactile perception in congenitally blind individuals. Neural Plasticity
2012: 1-8 doi: 10.1155/2012/720278.

Levinson, S.C. (2000) Presumptive meanings: The theory of generalized conversational

implicature. Cambridge, MA: MIT Press.

Lingnau, A., Strnad, L., He, C., Fabbri, S., Han, Z., Bi, Y. & Caramazza, A. (2014)
Cross-modal plasticity preserves functional specialization in posterior parietal cortex.
Cerebral Cortex 24: 541-549.

Liu, L. & Kager, R. (2016) Enhanced music sensitivity in 9-month-old bilingual infants.
Cognitive Processing 2016: 1-11 doi: 10.1007/s10339-016-0780-7.

Luria, A.R., Tsvetkova, L.S. & Futer, D.S. (1965) Aphasia in a composer (V.G.
Shebalin). Journal of the Neurological Sciences 2(3): 288-292.

MacDermot, K.D., Bonora, E., Sykes, N. Coupe, A.M., Lai, C.S.L., Vernes, S.C.,
Vargha-Khadem, F., McKenzie, F., Smith, R.L., Monaco, A.P. & Fisher, S.E. (2005)
Identification of FOXP2 truncation as a novel cause of developmental speech and
language deficits. American Journal of Human Genetics 76(6): 1074-1080.

MacNeilage, P.F. (1998) The frame/content theory of evolution of speech production.

Behavioral and Brain Sciences 21(4): 499-511; discussion 511-546.

Maess, B., Koelsch, S., Gunter, T.C. & Friederici, A.D. (2001) Musical syntax is
processed in Broca's area: An MEG study. Nature Neuroscience 4: 540-545.

Mahon, B.Z. & Cantlon, J.F. (2011) The specialization of function: Cognitive and neural
perspectives. Cognitive Neuropsychology 28(3-4): 147-155.

Malle, B.F. (2002) The relation between language and theory of mind in development
and evolution. In: The evolution of language out of pre-language, eds. T. Givón & B.
Malle, pp. 265-284. Philadelphia: John Benjamins.

Mameli, M. & Bateson, P. (2006) Innateness and the sciences. Biology and Philosophy 21:
155-188.

⎯⎯(2011) An evaluation of the concept of innateness. Philosophical Transactions of the

Royal Society B 366: 436-443.

Mameli, M. & Papineau, D. (2006) The new nativism: A commentary on Gary Marcus’s
The birth of the mind. Biology and Philosophy 21: 559-573.

Mampe, B., Friederici, A.D., Christophe, A. & Wermke, K. (2009) Newborns’ cry
melody is shaped by their native language. Current Biology 19(23): 1994-1997.

Marcus, G. (2004) The birth of the mind: How a tiny number of genes creates the
complexities of human thought. New York: Basic Books.
195
Marr, D. (1976) Early processing of visual information. Philosophical Transactions of the
Royal Society B 275: 483-524.

⎯⎯(2010 [1982]) Vision: A computational investigation into the human representation

and processing of visual information. Cambridge, MA: MIT Press.

Martin, A., Haxby, J.V., Lalonde, F.M., Wiggs, C.L. & Ungerleider, L.G. (1995)
Discrete cortical regions associated with knowledge of color and knowledge of action.
Science 270: 102-105.

Martin, A., Ungerleider, L.G. & Haxby, J.V. (2000) Category-specificity and the brain:
The sensorymotor model of semantic representations of objects. In: The new cognitive
neurosciences, 2nd edition, ed. M.S. Gazzaniga, pp. 1023-1036. Cambridge, MA, MIT
Press.

Martin, A., Wiggs, C.L., Ungerleider, L.G. & Haxby, J.V. (1996) Neural correlates of
category-specific knowledge. Nature 379(6566): 649-652.

Mather, M., Cacioppo, J.T. & Kanwisher, N. (2013) How fMRI can inform cognitive
theories. Perspectives on Psychological Science 8(1): 108-113.

McGeer, V. (2007) Why neuroscience matters to cognitive neuropsychology. Synthese

159: 347-371.

McGilvray, J. (2014) Chomsky: Language, mind, politics. 2nd ed. Cambridge: Polity
Press.

Melchner, L., Pallas, S.L. & Sur, M. (2000) Visual behaviour mediated by retinal
projections directed to the auditory pathway. Nature 404(6780): 871-876.

Merabet, L.B. & Pascual-Leone, A. (2010) Neural organization following sensory loss:
The opportunity of change. Nature Reviews Neuroscience 11: 44-52.

Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A.A., Veness, J., Bellemare, M.G., Graves,
A., Riedmiller, M., Fidjeland, A.K., Ostrovski, G., Petersen, S., Beattie, C., Sadik, A.,
Antonoglou, I., King, H., Kumaran, D., Wierstra, D., Legg S. & Hassabis, D. (2015)
Human-level control through deep reinforcement learning. Nature 518: 529-533.

Mountcastle, V.B. (1957) Modality and topographic properties of single neurons of cat’s
somatic sensory cortex. Journal of Neurophysiology 20(4): 408–434.

⎯⎯(1978) An organizing principle for cerebral function: The unit module and the
distributed system. In: The Mindful Brain, eds. G. Edelman & V.B. Mountcastle, pp. 7-
50. Cambridge, MA: MIT Press.

⎯⎯(1997) The columnar organization of the neocortex. Brain 120: 701-722.

196
Müller, R.-A. & Basho, S. (2004) Are nonlinguistic functions in “Broca’s area”
prerequisites for language acquisition? fMRI findings from an ontogenetic viewpoint.
Brain and Language 89(2): 329-336.

Neville, H.J. & Bavelier, D. (2001) Specificity of developmental neuroplasticity in

humans: Evidence from sensory deprivation and altered language experience. In: Toward
a theory of neuroplasticity, eds. C.A. Shaw & J.C. McEachern, pp. 261-274. Philadelphia:
Psychology Press.

Nishitani, N., Schürmann, M., Amunts, K. & Hari, R. (2005) Broca’s region: From
action to language. Physiology 20: 60-69.

Noppeney, U. (2007) The effects of visual deprivation on structural and functional

organization of the human brain. Neuroscience and Biobehavioral Reviews 31: 1169-1180.

Ohlsson, S. (1994) Representational change, generality versus specificity, and nature

versus nurture: Perennial issues in cognitive research. Behavioral and Brain Sciences
17(4): 724-725.

O’Neill, E. (2015) Relativizing innateness: Innateness as the insensitivity of the

appearance of a trait with respect to specified environmental variation. Biology and
Philosophy 30: 211-225.

O’Reilly, R.C. (1998) Six principles for biologically-based computational models of

cortical cognition. Trends in Cognitive Sciences 2(11): 455-462.

O’Reilly, R.C., Munakata, Y., Frank, M.J., Hazy, T.E. & Contributors (2012)
Computational cognitive neuroscience. 1st ed. Wiki Book. <http:// ccnbook.
colorado.edu>

Pascual-Leone, A., Amedi, A., Fregni, F. & Merabet, L.B. (2005) The plastic human
brain cortex. Annual Review of Neuroscience 28: 377-401.

Pascual-Leone, A. & Hamilton, R. (2001) The metamodal organization of the brain.

Progress in Brain Research 134: 427-445.

Pascual-Leone, A. & Torres, F. (1993) Plasticity of the sensorimotor cortex

representation of the reading finger in Braille readers. Brain 116: 39-52.

Pasqualotto, A. (2016) Multisensory integration substantiates distributed and overlapping

neural networks. Behavioral and Brain Sciences 39: 20-21.

Pessoa, L. (2016) Beyond disjoint brain networks: Overlapping networks for cognition
and emotion. Behavioral and Brain Sciences 39: 22-24.

Peretz, I. & Coltheart, M. (2003) Modularity of music processing. Nature Neuroscience

6: 688-691.

Petrov, A.A., Jilk, D.J. & O’Reilly, R.C. (2010) The Leabra architecture: Specialization
without modularity. Behavioral and Brain Sciences 33(4): 286-287.
197
Piccinini, G. & Craver, C. (2011) Integrating psychology and neuroscience: Functional
analyses as mechanism sketches. Synthese 183: 283-311.

Pinker, S. (1994) The language instinct. London: Penguin/Folio.

⎯⎯(1997) How the mind works. New York: Norton.

Pinker, S. & Jackendoff, R. (2005) The faculty of language: What’s special about it?
Cognition 95: 201-236.

Plaut, D.C. (1995) Double dissociation without modularity: Evidence from connectionist
neuropsychology. Journal of Clinical and Experimental Psychology 17(2): 291-321.

Poeppel, D. (2001) Pure word deafness and the bilateral processing of the speech code.
Cognitive Science 21(5): 679-693.

⎯⎯(2015) The neurobiology of language. In: The future of the brain, eds. G. Marcus &
J. Freeman, pp. 139-148. Princeton: Princeton University Press.

Poldrack, R.A. (2010) Mapping mental function to brain structure: How can cognitive
neuroimaging succeed? Perspectives on Psychological Science 5(6): 753-761.

Poldrack R.A., Halchenko, Y.O. & Hanson, S.J. (2009) Decoding the large-scale
structure of brain function by classifying mental states across individuals. Psychological
Science 20(11): 1364-1372.

Polger, T.W. (2008) Two confusions concerning multiple realization. Philosophy of

Science 75(5): 537-547.

⎯⎯(2009) Evaluating the evidence for multiple realization. Synthese 167(3): 457-472.

Polger, T.W. & Shapiro, L.A. (2008) Understanding the dimensions of realization.
Journal of Philosophy 105(4): 213-222.

⎯⎯(2016) The multiple realization book. Oxford: Oxford University Press.

Price C.J. & Friston K.J. (2005) Functional ontologies for cognition: The systematic
definition of structure and function. Cognitive Neuropsychology 22(3): 262-275.

Prinz, J.J. (2006) Is the mind really modular? In: Contemporary debates in cognitive
science, ed. R. Stainton, pp. 22-36. Oxford: Blackwell.

Ptito, M., Kupers, R., Lomber, S. & Pietrini, P. (2012) Sensory deprivation and brain
plasticity. Neural Plasticity 2012: 1-2 doi: 10.1155/2012/810370.

Ptito, M., Matteau, I., Zhi Wang, A., Paulson, O.B., Siebner, H.R. & Kupers, R. (2012)
Crossmodal recruitment of the ventral visual stream in congenital blindness. Neural
Plasticity 2012: 1-9 doi: 10.1155/2012/304045.

198
Pullum, G.K. & Scholz, B.C. (2002) Empirical assessment of stimulus poverty
arguments. Linguistic Review 19: 9-50.

Pulvermüller, F. (2005) Brain mechanisms linking language and action. Nature Reviews
Neuroscience 6: 576-582.

Pulvermüller, F. & Fadiga, L. (2010) Active perception: Sensorimotor circuits as a

cortical basis for language. Nature Reviews Neuroscience 11: 351-360.

Putnam, H. (1967) Psychological predicates. In: Art, mind, and religion, eds. W. Capitan
& D. Merrill, pp. 37-48. Pittsburgh: University of Pittsburgh Press.

Quartz, S.R. & Sejnowski, T.J. (1994) Neural evidence for constructivist principles in
development. Behavioral and Brain Sciences 17(4): 725-726.

Ramus, F. & Fisher, S.E. (2009) Genetics of Language. In: The cognitive neurosciences,
4th edition, ed. M.S. Gazzaniga, pp. 855-872. Cambridge, MA: MIT Press.

Rauschecker, J.P. (2001) Developmental neuroplasticity within and across sensory

modalities. In: Toward a theory of neuroplasticity, eds. C.A. Shaw & J.C. McEachern,
pp. 244-260. Philadelphia: Psychology Press.

Renier, L.A., Anurova, I., De Volder, A.G., Carlson, S., VanMeter, J. & Rauschecker,
J.P. (2010) Preserved functional specialization for spatial processing in the middle
occipital gyrus of the early blind. Neuron 68(1): 138-148.

Richerson, P.J. & Boyd, R. (2005) Not by genes alone: How culture transformed human
evolution. Chicago: University of Chicago Press.

Ritchie, J.B. & Carruthers, P. (2010) Massive modularity is consistent with most forms of
neural reuse. Behavioral and Brain Sciences 33(4): 289-290.

Robbins, P. (2009) Modularity of mind. In: The Stanford Encyclopedia of Philosophy,

summer 2010, ed. E.N. Zalta. <http://plato.stanford.edu/archives/sum2010/entries/
modularity-mind/>

Rockland, K.S. (2010) Five points on columns. Frontiers in Neuroanatomy 4(6): 1-10.

Rose, J.K. & Rankin, C.H. (2001) Behavioral, neural circuit and genetic analyses of
habituation in C. elegans. In: Toward a theory of neuroplasticity, eds. C.A. Shaw & J.C.
McEachern, pp. 176-192. Philadelphia: Psychology Press.

Rowland, D.C. & Moser, M.B. (2014) From cortical modules to memories. Current
Opinion in Neurobiology 24: 22-27.

Sadato, N., Pascual-Leone, A., Grafman, J., Ibañez, V., Deiber, M.P., Dold, G. &
Hallett, M. (1996) Activation of the primary visual cortex by Braille reading in blind
subjects. Nature 380(6574): 526-528.

199
Saitoe, M. & Tully, T. (2001) Making connections between developmental and
behavioral plasticity in Drosophila. In: Toward a theory of neuroplasticity, eds. C.A.
Shaw & J.C. McEachern, pp. 193-220. Philadelphia: Psychology Press.

Sanes, J.R. & Jessell, T.M. (2013) Experience and the refinement of synaptic
connections. In: Principles of neural science, eds. E.R. Kandel, J.H. Schwartz, T.M.
Jessell, S.A. Siegelbaum & A.J. Hudspeth, pp. 1259-1283. New York: McGraw-Hill.

Scholl, B.J. (1997) Neural constraints on cognitive modularity? Behavioral and Brain
Sciences 20(4): 575-576.

Scholz, B.C. & Pullum, G.K. (2002) Searching for arguments to support linguistic
nativism. Linguistic Review 19: 185-223.

Schreiweis, C., Bornschein, U., Burguière, E., Kerimoglu, C., Schreiter, S., Dannemann,
M., Goyal, S., Rea, E., French, C.A., Puliyadi, R., Groszer, M., Fisher, S.E., Mundry,
R., Winter, C., Hevers, W., Pääbo, S., Enard, W. & Graybiel, A.M. (2014) Humanized
Foxp2 accelerates learning by enhancing transitions from declarative to procedural
performance. Proceedings of the National Academy of Sciences of the United States of
America 111(39): 14253-14258.

Shapiro, L.A. (2000) Multiple realizations. Journal of Philosophy 97(12): 635-654.

⎯⎯(2004) The mind incarnate. Cambridge, MA: MIT Press.

⎯⎯(2008) How to test for multiple realization. Philosophy of Science 75(5): 514-525.

Shapiro, L.A. & Polger, T.W. (2012) Identity, variability, and multiple realization in the
special sciences. In: New perspectives on type identity: The mental and the physical, eds.
S. Gozzano & C.S. Hill, pp. 264-286. Cambridge: Cambridge University Press.

Shaw, C.A. & McEachern, J.C., eds. (2001) Toward a theory of neuroplasticity.
Philadelphia: Psychology Press.

Sharma, J., Angelucci, A. & Sur, M. (2000) Induction of visual orientation modules in
auditory cortex. Nature 404(6780): 841-847.

Silberstein, M. (2016) The implications of neural reuse for the future of both cognitive
neuroscience and folk psychology. Behavioral and Brain Sciences 39: 27-29.

Smith, K. & Kirby, S. (2008) Cultural evolution: Implications for understanding the
human language faculty and its evolution. Philosophical Transactions of the Royal
Society B 363: 3591-3603.

Sperber, D. (1994) The modularity of thought and the epidemiology of representations.

In: Mapping the mind, eds. L.A. Hirschfield & S.A. Gelman, pp. 39-67. Cambridge:
Cambridge University Press.

⎯⎯(2002) In defense of massive modularity. In: Language, brain, and cognitive

development, ed. I. Dupoux, pp. 47-57. Cambridge, MA: MIT Press.
200
Sporns, O. (2015) Network neuroscience. In: The future of the brain, eds. G. Marcus &
J. Freeman, pp. 90-99. Princeton: Princeton University Press.

Stanley, M.L. & De Brigard, F. (2016) Modularity in network neuroscience and neural
reuse. Behavioral and Brain Sciences 39: 29-31.

Sterelny, K. (2006) Language, modularity, and evolution. In: Teleosemantics, eds. G.

Macdonald & D. Papineau, pp. 23-41. Oxford: Oxford University Press.

⎯⎯(2012) The evolved apprentice. Cambridge, MA: MIT Press.

Sternberg, S. (2011) Modular processes in mind and brain. Cognitive Neuropsychology

28(3-4): 156-208.

Striem-Amit, E., Dakwar, O., Reich, L. & Amedi, A. (2012) The large-scale organization
of “visual” streams emerges without visual experience. Cerebral Cortex 22(7): 1698-1709.

Striem-Amit, E. & Amedi, A. (2014) Visual cortex extrastriate body-selective area

activation in congenitally blind people “seeing” by using sounds. Current Biology 24: 1-6.

Suddendorf, T. (2013) The gap: The science of what separates us from the animals. New
York: Basic Books.

Teyler, T.J. (2001) LTP and the superfamily of synaptic plasticities. In: Toward a theory
of neuroplasticity, eds. C.A. Shaw & J.C. McEachern, pp. 101-117. Philadelphia:
Psychology Press.

Thoenissen, D., Zilles, K. & Toni, I. (2002) Differential involvement of parietal and
precentral regions in movement preparation and motor intention. Journal of
Neuroscience 22: 9024-9034.

van Gelder, T. (1995) What might cognition be, if not computation? Journal of
Philosophy 92(7): 345-381.

Varley, R.A., Klessinger, N.J.C., Romanowski, C.A.J. & Siegal, M. (2005) Agrammatic
but numerate. Proceedings of the National Academy of Sciences of the United States of
America 102: 3519-3524.

Verhage, M., Maia, A.S., Plomp, J.J., Brussaard, A.B., Heeroma, J.H., Vermeer, H.,
Toonen, R.F., Hammer, R.E., van den Berg, T.K., Missler, M., Geuze, H.J. & Südhof,
T.C. (2000) Synaptic assembly of the brain in the absence of neurotransmitter secretion.
Science 287(5454): 864-869.

Waddington, C.H. (1953) Genetic assimilation of an acquired character. Evolution 17:

118-126.

⎯⎯(1955) On a case of quantitative variation on either side of the wild type. Molecular
Genetics and Genomics 87(2): 208-228.

201
Wernicke, C. (1908) The symptom-complex of aphasia. In: Diseases of the nervous
system, ed. A. Church, pp. 265-324. New York: Appleton.

Williams, L.E. & Bargh, J.A. (2008a) Experiencing physical warmth promotes
interpersonal warmth. Science 322: 606-607.

⎯⎯(2008b) Keeping one’s distance: The influence of spatial distance cues on affect and
evaluation. Psychological Science 19: 302-308.

Wolbers, T., Zahorik, P. & Giudice, N.A. (2011) Decoding the direction of auditory
motion in blind humans. NeuroImage 56(2): 681-687.

Wolpert, L. (2011) Developmental biology: A very short introduction. New York: Oxford
University Press.

Zador, A. (2015) The connectome as a DNA sequencing problem. In: The future of the
brain, eds. G. Marcus & J. Freeman, pp. 40-49. Princeton: Princeton University Press.

Zerilli, J. (2014) A minimalist framework for comparative psychology. Biology and

Philosophy 29(6): 897-904.

⎯⎯(2017a) Against the “system” module. Philosophical Psychology 30(3): 231-246.

⎯⎯(2017b) Multiple realization and the commensurability of taxonomies. Synthese (in

press).

202