Modularity and Domain Specificity

H. CLARK BARRETT
University of California, Los Angeles, United States

Modularity refers to the degree to which a system (e.g., a network of people, ideas, or
neurons) can be decomposed into semiautonomous parts or units. Domain specificity
refers to the degree to which a particular process is modified as a function of what
it is operating on. When applied to the brain and mind, these concepts offer insight
into how the apparently seamless whole of thought and behavior is constructed via the
interaction of underlying processes and mechanisms.
In anthropology, modularity and domain specificity can be and are applied differ-
ently in different areas of research. For example, one can speak of the modularity of
institutions, or the modularity of exchange networks. However, the concepts of modu-
larity and domain specificity are most commonly used in the sciences of cognition and
mind, including cognitive anthropology, evolutionary psychology, and neuroscience,
where they are specifically applied to the mind and brain. In this literature, modularity
refers to the decomposability of brain processes and structures into semiautonomous
components and domain specificity refers to ways in which mental processes differ as
a function of the information being processed. Because both concern how processes in
the mind are structured, modularity and domain specificity are intimately related.

Modularity

Classical modularity: Modularity as encapsulation

The earliest and most widespread definition of modularity in the cognitive sciences
was introduced by philosopher Jerry Fodor in his book The Modularity of Mind (1983).
Fodor offered a checklist of features he proposed to be characteristic of modules,
including domain specificity, informational encapsulation, obligatory firing, fast speed,
shallow outputs, limited accessibility, characteristic ontogeny, and fixed neural archi-
tecture. He argued that modularity was a matter of degree and that different systems
in the mind might be modular to varying extents and possess different combinations
of features. However, many, including Fodor himself, came to regard encapsulation as
the key feature of modularity. On this “classical” view of modularity, then, modularity
is mostly defined in terms of the computational autonomy of the processes involved.
Encapsulation refers to the strict independence of processes occurring within a
module from processes occurring outside of the module. Modules, on this view, are
like subroutines in a computer program, or objects in object-oriented programming
languages. They communicate with other processes only via rigidly defined interfaces

The International Encyclopedia of Anthropology. Edited by Hilary Callan.

© 2018 John Wiley & Sons, Ltd. Published 2018 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118924396.wbiea1740
2 MODUL A R ITY A ND D O M A I N S P E CI F I CI T Y

through which inputs are passed to the module and outputs are passed out of it. The
analogy to computer programs here is explicit, as classical modularity is part of the
“computational theory of mind” or CTM, which seeks to understand mental processes
as formal processes of computation.
While the classical view of modularity has been extremely influential, it has also come
under mounting criticism as a general model for cognition, including by Fodor him-
self. The reasons are both theoretical and empirical. Theoretically, classically modular
systems seem ill equipped to account for mental processes that are heavily interactive
and contextually adjustable. Models involving parallel distributed computation within
a network of interacting systems seem better suited for handling cognitive problems of
this kind. Empirically, much of cognition appears to be of the latter sort, highly flexible
and context sensitive. To the extent that classical modules are akin to cognitive reflexes,
then, they seem ill-suited to account for the more flexible and less reflex-like aspects
of cognition. In addition, the classical model is committed to a two-layer architecture,
with a layer of “peripheral” modules corresponding to sensory and motor systems sur-
rounding a module-free “central” system. Studies of network modularity suggest that
this architecture is probably incorrect. The structure of the brain is better described as
hierarchically modular, composed of nested and interacting modules all the way up,
rather than being composed of a modular and a nonmodular layer.

Network modularity: Modularity as structured connectivity

In network theory, modularity is defined mathematically. It is an empirically measur-
able property of networks. Networks are systems that can be represented formally as
graphs composed of nodes connected by edges. Thus, network theory is a formalism
that can be applied to a wide range of entities, from genes to people to scientific papers.
In this literature, modularity is conceptualized as the relative “withinness,” or local-
ity of connections, in a network. Highly modular networks are characterized by high
within-region connectivity relative to between-region connectivity. A useful formaliza-
tion defines network modularity as the fraction of edges that fall within a collection of
nodes compared with what would be expected at random. In network theory as in clas-
sical modularity, then, modularity is a matter of degree but has a more precise formal
operationalization in the case of networks.
Recently, neuroscientists have begun to apply network theory to patterns of neu-
ral connectivity within the brain, sometimes called the “connectome” (Sporns 2012).
The primary aim of this work is to produce a detailed map of the human connectome
and ultimately to link this structural description to function—that is, cognition and
behavior. While this work is still in its infancy, two broad conclusions can already be
drawn. First, the brain is highly modular in the network sense: of all possible connec-
tions that could be formed across the brain’s vast network, only a small fraction are, and
local connections are much more common than long-distance ones when compared
with what would be expected by chance. Second, the brain’s modularity is hierarchical:
modularity occurs at multiple scales, with modules at one scale organized modularly at
higher scales. Hierarchical modularity of this kind is typical of the design of organ-
isms, which exhibit modularity at scales ranging from biochemical networks at the
 MODUL A R ITY A ND D O M A I N S P E CI F I CI T Y 3

lowest levels of the hierarchy to networks of organs and other function structures at the
whole-organism level. Indeed, the brain and nervous system, which together comprise a
functional module within animals, span the full range of this hierarchy, from biochem-
icals such as hormones and neurotransmitters up to large-scale functional complexes
such as the limbic system and the cortex. From this perspective, then, the discovery of
substantial modularity in the brain by connectomics researchers is not surprising.

Evolutionary modularity: Modularity as evolvability

In evolutionary developmental biology (evo-devo), modularity is again conceptualized
in terms of decomposability of processes but here the relevant aspect of decompos-
ability is in terms of developmental ingredients that can be acted on by evolutionary
processes. A key concept in this literature is evolvability, which refers to the possible
pathways the evolutionary process can take as a function of combinations made pos-
sible by the genetic and developmental systems that give rise to organismic variation
(Wagner and Altenberg 1996). On this view, the evolutionary modularity of a devel-
opmental system refers to the degree to which different aspects of the system can be
acted on independently by evolutionary processes such as natural selection. Develop-
mental processes are evolutionarily modular to the extent that they are separable and
therefore can be altered independently or semi-independently over evolutionary time.
Important to this view of modularity is that developmental modules may cross-cut phe-
notypic modules. For example, a single underlying developmental module could give
rise to multiple phenotypic modules, such as hair follicles, limbs, or brain regions. Con-
versely, multiple developmental modules could, and probably usually do, interact in the
development of phenotypic traits. A primary concern of this literature is to understand
how the evolutionary process shapes whole systems of interacting parts (organisms) as
a function of how those parts interact in a modular hierarchy.
In biology, the evo-devo perspective on modularity contrasts with earlier views of
organisms as assemblages of more or less independent traits, each controlled by its own
genes. Early research in evolutionary developmental biology quickly overturned the
idea that genes and phenotypes stand in a one-to-one relationship to each other such
that “modules” at the level of the phenotype correspond to “modules” at the level of the
genome. Instead, the genome-to-phenome mapping relationship is complex and biolo-
gists are only beginning to understand how evolutionary change occurs by alterations
to underlying developmental modules that have diverse phenotypic effects. However,
evolutionary modularity is clearly important for understanding the network modular-
ity of the brain and the organization of mental processes that the brain’s structure gives
rise to. The evolutionary changes that have enabled human capacities of cognition, cul-
ture, and behavior must have happened, at least in part, via modifications of underlying
developmental modules that build the brain.

Hierarchical modularity as a unifying concept

Modularity is defined differently in different fields, yet these definitions need not
be mutually exclusive. Network modularity, for example, is entirely compatible with
4 MODUL A R ITY A ND D O M A I N S P E CI F I CI T Y

evolutionary modularity, provided that one understands that networks of synaptic

connections are part of the developed phenotype whereas the network of resources
that build the connectome is a developmental network composed of as-yet-unknown
developmental modules. In turn, cognitive modularity, if defined as the (partial)
decomposability of cognition into underlying processes, must depend at least in part
on the organization of the connectome.
To the extent that it implies modularity at many levels and of different types, the idea
of hierarchical modularity has the capacity to unify these different, diverse concepts of
modularity, both in terms of modularity of developmental processes and modularity
of brain structure. The processes that build the brain and give rise to the mind oper-
ate on many timescales, from the evolutionary timescale (macro) to the developmental
timescale (meso) to the timescale of neural processing (micro). There may be modular-
ity of processes at all of these timescales simultaneously and the modularity that occurs
at some timescales (e.g., neural processing) may depend on modularity at other scales
(e.g., developmental). Thus, the modularity of brain processes may be temporally and
functionally hierarchical. In addition, hierarchical modularity is likely to occur at many
levels of brain structure, from the level of gene expression networks and their biochem-
ical products to the level of tissue organization within a brain region, all the way up to
the macro level of brain organization.
For most anthropologists, the relevance of modularity (or lack thereof) lies in its
implications for cognition, culture, and behavior. Here, the related concept of domain
specificity becomes important.

Domain specificity

Like modularity, domain specificity has been used differently by different authors. In
the usage that most naturally follows from the classical modularity concept, domain
specificity is treated as a binary, either/or category. Under this conception, psychological
processes are either domain specific or domain general, with domain-specific processes
corresponding to the activity of modular, peripheral systems and domain-general pro-
cesses corresponding to the activity of nonmodular central systems. This binary oppo-
sition is common in currently popular “dual systems” views of the mind in psychology.
An alternative conception of domain specificity treats it as a matter of degree, much
as modularity is treated in network theory. On this view, the domain specificity of a
process may vary as a function of context and/or the level of a hierarchy that one is con-
sidering. Precise formal definitions of domain specificity are possible given sufficiently
well-defined concepts of process and domain for the case in question. One can then
define the domain specificity of a process as the degree to which its operations vary
across domains. A perfectly domain-specific process is one that operates only in a sin-
gle domain and in no others. A perfectly domain-general process is one that operates
identically across all domains. In between are processes that exhibit degrees of domain
specificity, meaning that their operations are modulated or modified depending on the
domain in which they are operating. Most processes of “higher-level” cognition, such as
 MODUL A R ITY A ND D O M A I N S P E CI F I CI T Y 5

reasoning and language, probably exhibit some intermediate level of domain specificity
when defined in this way.
In the cognitive sciences, the notion of domain specificity originated as part of
the so-called cognitive turn which arose in opposition to behaviorist psychology in
the 1950s and 1960s. In its strictest form, behaviorism sought to banish discussion
of the mind and mental processes from psychology on the grounds that processes
inside the mind are unobservable and therefore outside the purview of scientific
investigation. Thus, behaviorists limited their study to relationships between “stimuli,”
observable events in the external world, and “responses,” behaviors produced puta-
tively in response to those stimuli. An important tenet of behaviorism was the notion of
equipotentiality: any response could be paired equally well with any stimulus, given the
proper conditioning. The equipotentiality assumption was challenged by the finding
that rats more easily learn to associate nausea with the taste of a food they have eaten
than with other temporally associated experiences such as electric shock (Garcia and
Koelling 1966). These so-called learned food aversions are examples of domain-specific
learning: they are a form of classical conditioning, which occurs across domains, but the
conditioning is “prepared” to rapidly associate nausea with taste in the domain of food.
The term “domain specificity” first became widely used in the cognitive development
literature in the late 1970s and early 1980s. That literature grew out of psychologist
Jean Piaget’s “stage” theories of cognitive development, which held that development
proceeded in transitions between stages that operated across cognitive domains: the
sensorimotor, preoperational, concrete operations, and formal operations stages. While
Piaget knew that developmental transitions could differ in their timing across domains,
a phenomenon he called “décalage,” stage transitions were given primacy in his theory.
Faced with data suggesting that infants develop sophisticated abilities early in some
domains and late in others, some cognitive developmentalists suggested that Piaget had
it backward: development is fundamentally organized not around stages but around
“core” domains such as the understanding of physical objects, intentional agents, and
numerosity (Carey 2009).

Varieties of domain specificity

Stemming from its popularization by developmental psychologists studying “core”
domains, domain specificity is most commonly associated in psychology with “content”
domains—that is, domains defined by the meaning or semantics of what is being oper-
ated on. Examples of domains of this kind are the domains of animals, foods, artifacts,
and kin. Evolutionary psychology in particular is known for proposing a host of mod-
ules specialized for content domains of this kind. In keeping with this proposal, there
is indeed neuropsychological evidence for brain processes specialized to particular
kinds of content, such as faces, places, artifacts, and mental states (Kanwisher 2010).
While this notion of content domains has proven useful, other, broader definitions
of “domain” are possible and will probably prove necessary in order to capture the
full scope of the organization of mental processes. In mathematics, for example, the
domain of a function is the set of input values for which the function is defined. This
is a very broad definition but one that fits naturally with a view of mental processes
6 MODUL A R ITY A ND D O M A I N S P E CI F I CI T Y

as computations (CTM). On this view, whatever computational process we might

examine will, by definition, have a domain. However, evolutionary psychologists
distinguish between two functional senses of domain. The actual domain of a compu-
tational process is the set of inputs for which the process can or will produce outputs
whereas the proper domain is the set of inputs for which the process has been selected
to produce outputs. This is roughly akin to the distinction between “selected effect”
and “causal role” functions in biology: the selected effect function of a biological
trait is the function it was selected to carry out (in the past) whereas the causal role
function of a trait includes all of the functions it can in fact perform, whether or
not it has been selected to do so (e.g., a causal role function of the heart is to make
noise though it has not been specifically selected to do this). For domains, the scope
of the proper domain always exceeds the scope of the actual domain because the set
of inputs a mechanism can process (the actual domain) always includes the set of
inputs it was selected to process (the proper domain) plus an additional set of inputs
that it is able to process by virtue of its computational properties, even though it
was not selected to do so. Thus, some domain-specific processing in the mind will
be “by-product” processing: even evolutionarily novel inputs may be processed in a
domain-specific way if they are within the actual domain of evolved domain-specific
processes.
Technically, if one defines the inputs to a function or mechanism as the content
that the mechanism processes, then all domains are content domains. However, it is
important to keep in mind that the content domains around which mental processes
are organized might not always correspond to domains of meaning or semantics in
natural languages. For example, the visual system contains processes whose proper
domain is the set of all edges with a specific orientation on the retina, a category that
does not map to a semantic domain in natural languages. At so-called “higher” levels
of cognition, too, such as moral reasoning or causal inference, it is important to keep in
mind that the “joints” at which the mind carves domains might not always correspond
to joints that are recognized in natural languages or folk theories. Because intuitions
shape research, it is presently difficult to say how much the brain’s domain organization
departs from intuitions about “content” types. The research area known as cognitive
phenomics takes a data-driven approach to this question by examining the correla-
tion structure of performance on various tasks and associated brain activity as revealed
through brain mapping using meta-analysis of multiple studies to tease apart the rela-
tionship between specific tasks (e.g., cognitive load tasks) and putative psychological
constructs (e.g., working memory), resulting in a searchable “cognitive atlas” (Poldrack
et al. 2011). Thus, this approach has the potential to reveal the domain structure of
cognitive processes in a bottom-up manner.

Implications for anthropology

For anthropologists studying the varieties of human behavior and cognition across
cultures, the organization of the brain into modules and the organization of cognitive
processes into domains has potentially important implications. Some anthropologists
 MODUL A R ITY A ND D O M A I N S P E CI F I CI T Y 7

might assume, implicitly, that human thought and behavior is infinitely malleable.
Indeed, there is no question that a catalog of things that humans could in principle
think or do would be limitless. However, to say that a combinatorial system can
produce an infinite array of outputs is not to say that those outputs cover all areas of the
possibility space; the rules of chess produce infinitely many possible game sequences
but each game is highly constrained by the rules. Similarly, if human thought and
activity is generated by evolved mental structures, the combinations produced may be
both infinite and yet limited to those the underlying system is able to generate.
Even if the computational theory of mind is correct (and if one allows a broad enough
definition of computation there are few alternatives), this does not imply that thought
and behavior are generated unidirectionally by the innate structure of the system alone.
Humans have culture and learning, which are processes of information acquisition. This
means that the domain specificity of mental processes may play a role in both infor-
mation acquisition (inputs) and behavioral production (outputs). However, there exist
different conceptions of just what such a role might be.

Conceptions of domain specificity in cultural transmission

There is a range of opinions regarding the role that domain-specific processes are likely
to play in shaping culture and mental activity. For many anthropologists, the role that
mental processes play in shaping culture is of relatively little interest and many scholars
might doubt that domain-specific processes shape culture at all. Among those that do
entertain the possibility, there are two main schools of thought.
A view that is popular among evolutionary psychologists is that domain-specific
mechanisms place strong constraints on what information can be acquired and repre-
sented by the mind. On this view, domain-specific mechanisms act somewhat like filters
and the role that domain-specific mechanisms play is likened to “attractors” in dynami-
cal systems: while representations are transmitted from mind to mind, the distribution
of these representations in a population of minds is strongly constrained by the proper-
ties of domain-specific mechanisms. This is sometimes called a “cultural attractor” view
(Sperber 1996).
Another perspective holds that domain-specific mechanisms are more akin to
transmission biases than to filters. On this view, mechanisms nudge the probability
of acquiring information in one direction or the other but do not strictly rule acqui-
sition in or out. Examples of transmission biases include prestige bias (a tendency
to favor information from higher prestige individuals), conformity bias (a tendency to
favor majority variants of information), and various content biases (a tendency to favor
information about, e.g., danger, sex, and the like). This model of cultural transmission is
most closely associated with culture–gene coevolution theorists, who tend to downplay
the role of domain specificity in information transmission (Richerson & Boyd 2005).
However, theorists in this field tend to adopt the “content” view of domains rather than
the broader, mathematically defined concept of domain offered earlier. Because all of
the biases described here have input conditions, all are domain specific according to
the broad definition.
8 MODUL A R ITY A ND D O M A I N S P E CI F I CI T Y

To some extent, this difference of opinion is historical and based on theoretical dif-
ferences. In practice, relatively little is known, empirically, about the importance of
domain-specific mechanisms—or any mechanisms, for that matter—in structuring cul-
tural transmission. Moreover, the “attractor” and “bias” views are not, in fact, mutually
exclusive. Biases can shape basins of attraction in dynamical systems and biases also act
as probabilistic filters. One difference between these schools of thought surfaces in intu-
itions about the historical dynamics of culture, with the attractor school treating cultural
history as highly predetermined and the bias school treating it as largely unbounded
and chaotic. Additionally, culture–gene coevolution theorists point to strong historical
evidence of cumulative, or ratcheting, cultural evolution, in which cultural products
and representations, such as grammars and artifacts, ratchet upwards in complexity
over time. Theoretically, there is no conflict between the idea of cumulative cultural
evolution and domain specificity but theory has yet to clearly resolve the question of
what kinds of cognitive mechanisms make cumulative cultural evolution possible and
there have been few strong empirical tests that carefully tease apart different possible
mechanisms of cultural transmission.

Modularity and human nature

Debates about modularity and domain specificity in the literature have been riddled
with misconceptions, often stemming from the historical baggage associated with these
terms rather than internally consistent logic. For example, while modularity is often
associated with rigidity (an implication of the classical view), there is no reason why a
modular mind cannot be a flexible mind. Indeed, in the evo-devo literature, a primary
virtue of modularity is the flexibility and robustness of modular systems. Moreover, in
development, modular processes involving the interaction of diverse mechanisms may
lead to substantially variable outcomes under different developmental circumstances.
Thus, modularity and domain specificity do not mandate a heavily nativist view of
human nature and are consistent with substantial amounts of individual and cultural
variation.
To the extent that the mind’s modular organization has been shaped by the evolu-
tionary process, understanding it is likely to shed light on human nature and human
variability. For this reason, efforts to map the mind’s modular organization, whatever it
might turn out to be, should be of interest to anthropologists. However, new perspec-
tives on modularity arising from biology present a substantially different view of what
“human nature” might be than what the classical view of modularity suggested.
From a biological perspective, “human nature” and “human uniqueness” are not the
same thing. For example, many would consider parental love and nurturing of offspring
to be part of human nature but this is not a uniquely human trait. Indeed, most of what
we consider to be human nature is likely to be homologous with other species, with
a small portion consisting of traits that are uniquely derived in the human lineage.
Furthermore, human variation is not inconsistent with human nature. Virtually all
biological traits show variation within a species, yet species have characteristic traits.
Hummingbird beaks, for example, show variation within a species and yet beak
 MODUL A R ITY A ND D O M A I N S P E CI F I CI T Y 9

morphology is certainly part of the species-typical design of a given hummingbird

species.
When it comes to the modular organization of human brains, we know that human
brains evolved through descent with modification from ancestral primate brains and,
therefore, most of the modular organization of our brains is ancient and homologous
with that seen in other primates. Consequently, much of the domain-specific organiza-
tion of human thought may be homologous with that of other primates and mammals;
for example, there may be cognitive processes specialized to domains such as food, sex,
kinship, and cooperation.
Despite this deep homology, it is a mistake to believe that uniquely human aspects of
cognition, whatever those might be, evolved through the addition of one or a few new
modules to the ancestral primate brain, such that human thought consists of nonhuman
primate thought plus an additional domain or two (e.g., culture, language). Instead, evo-
lutionary developmental biology has shown that much evolutionary change proceeds
via evolutionary tweaks at many points in modular developmental systems. Small devel-
opmental changes can have large phenotypic effects, and just as the mapping between
genes and phenotypes is not one-to-one, neither is the mapping between alterations
in developmental modules (e.g., modules of gene expression) and resulting phenotypic
changes. Indeed, the evolutionary changes that have led to modern human cognition
almost certainly involve a complex mix of evolutionary changes at various levels of the
hierarchically modular organization of the brain. Phenomena such as culture and lan-
guage may involve some domain-specific processes but also almost certainly interact
with other processes in the mind and their evolution has likely caused alterations in
phylogenetically older brain structures and processes.
The move from classical to more biologically grounded, interactivist approaches
to modularity has the potential to create consilience between the complex models of
human cognition and behavior endorsed by anthropologists and evolutionary models
of human cognition. Adoption of a biological perspective on modularity invalidates
many prior critiques of modular accounts of the mind, such as the classical account.
However, it also means that the work of understanding how modular organization gives
rise to human nature is still in its infancy and is likely to be much more complicated
than older models of modularity suggested.

SEE ALSO: Brain and Culture; Cognition; Cognitive Development; Cultural Transmis-
sion; Ethnobiology and Cognition; Evolutionary Psychology; Mind; Relevance

REFERENCES AND FURTHER READING

Barrett, H. Clark. 2015. The Shape of Thought: How Mental Adaptations Evolve. New York: Oxford
University Press.
Carey, Susan. 2009. The Origin of Concepts. Cambridge, MA: MIT Press.
Fodor, Jerry. 1983. The Modularity of Mind. Cambridge, MA: MIT Press.
Garcia, John, and Robert A. Koelling. 1966. “Relation of Cue To Consequence in Avoidance
Learning.” Psychonomic Science 4: 123–24.
10 MODUL A R ITY A ND D O M A I N S P E CI F I CI T Y

Kanwisher, Nancy. 2010. “Functional Specificity in the Human Brain: A Window into the
Functional Architecture of the Mind.” Proceedings of the National Academy of Sciences 107:
11163–70.
Poldrack, Russell A., Aniket Kittur, Donald Kalar, Eric Miller, Christian Seppa, Yolanda Gil,
D. Stott Parker, Fred W. Sabb, and Robert M. Bilder. 2011. “The Cognitive Atlas: Toward
a Knowledge Foundation for Cognitive Neuroscience.” Frontiers in Neuroinformatics 5: 17.
doi:10.3389/fninf.2011.00017.
Richerson, Peter J., and Robert Boyd. 2005. Not By Genes Alone: How Culture Transformed
Human Evolution. Chicago: University of Chicago Press.
Sperber, Dan. 1996. Explaining Culture: A Naturalistic Approach. New York: Blackwell.
Sporns, Olaf. 2012. Discovering the Human Connectome. Cambridge, MA: MIT Press.
Tooby, John, Leda Cosmides, and Jerome Barkow, eds. 1992. The Adapted Mind: Evolutionary
Psychology and the Generation of Culture. New York: Oxford University Press.
Wagner, Gunter P., and Lee Altenberg. 1996. “Complex Adaptations and the Evolution of Evolv-
ability.” Evolution 50: 967–76.