Skip to main content
Social Sci LibreTexts

3.15: Modularity of Mind

  • Page ID
    35726
  • \( \newcommand{\vecs}[1]{\overset { \scriptstyle \rightharpoonup} {\mathbf{#1}} } \) \( \newcommand{\vecd}[1]{\overset{-\!-\!\rightharpoonup}{\vphantom{a}\smash {#1}}} \)\(\newcommand{\id}{\mathrm{id}}\) \( \newcommand{\Span}{\mathrm{span}}\) \( \newcommand{\kernel}{\mathrm{null}\,}\) \( \newcommand{\range}{\mathrm{range}\,}\) \( \newcommand{\RealPart}{\mathrm{Re}}\) \( \newcommand{\ImaginaryPart}{\mathrm{Im}}\) \( \newcommand{\Argument}{\mathrm{Arg}}\) \( \newcommand{\norm}[1]{\| #1 \|}\) \( \newcommand{\inner}[2]{\langle #1, #2 \rangle}\) \( \newcommand{\Span}{\mathrm{span}}\) \(\newcommand{\id}{\mathrm{id}}\) \( \newcommand{\Span}{\mathrm{span}}\) \( \newcommand{\kernel}{\mathrm{null}\,}\) \( \newcommand{\range}{\mathrm{range}\,}\) \( \newcommand{\RealPart}{\mathrm{Re}}\) \( \newcommand{\ImaginaryPart}{\mathrm{Im}}\) \( \newcommand{\Argument}{\mathrm{Arg}}\) \( \newcommand{\norm}[1]{\| #1 \|}\) \( \newcommand{\inner}[2]{\langle #1, #2 \rangle}\) \( \newcommand{\Span}{\mathrm{span}}\)\(\newcommand{\AA}{\unicode[.8,0]{x212B}}\)

    Classical cognitive science assumes that cognition is computation, and endorses the physical symbol system hypothesis. As a result, it merges two theoretical positions that in the seventeenth century were thought to be in conflict. The first is Cartesian rationalism, the notion that the products of thought were rational conclusions drawn from the rule-governed manipulation of pre-existing ideas. The second is anti-Cartesian materialism, the notion that the processes of thought are carried out by physical mechanisms.

    The merging of rationalism and materialism has resulted in the modification of a third idea, innateness, which is central to both Cartesian philosophy and classical cognitive science. According to Descartes, the contents of some mental states were innate, and served as mental axioms that permitted the derivation of new content (Descartes, 1996, 2006). Variations of this claim can be found in classical cognitive science (Fodor, 1975). However, it is much more typical for classical cognitive science to claim innateness for the mechanisms that manipulate content, instead of claiming it for the content itself. According to classical cognitive science, it is the architecture that is innate.

    Innateness is but one property that can serve to constrain theories about the nature of the architecture (Newell, 1990). It is a powerful assumption that leads to particular predictions. If the architecture is innate, then it should be universal (i.e., shared by all humans), and it should develop in a systematic pattern that can be linked to biological development. These implications have guided a tremendous amount of research in linguistics over the last several decades (Jackendoff, 2002). However, innateness is but one constraint, and many radically different architectural proposals might all be consistent with it. What other constraints might be applied to narrow the field of potential architectures?

    Another constraining property is modularity (Fodor, 1983). Modularity is the claim that an information processor is not just one homogeneous system used to handle every information processing problem, but is instead a collection of specialpurpose processors, each of which is especially suited to deal with a narrower range of more specific problems. Modularity offers a general solution to what is known as the packing problem (Ballard, 1986).

    The packing problem is concerned with maximizing the computational power of a physical device with limited resources, such as a brain with a finite number of neurons and synapses. How does one pack the maximal computing power into a finite brain? Ballard (1986) argued that many different subsystems, each designed to deal with a limited range of computations, will be easier to fit into a finite package than will be a single general-purpose device that serves the same purpose as all of the subsystems.

    Of course, in order to enable a resource-limited system to solve the same class of problems as a universal machine, a compromise solution to the packing problem may be required. This is exactly the stance adopted by Fodor in his influential 1983 monograph The Modularity of Mind. Fodor imagined an information processor that used general central processing, which he called isotropic processes, operating on representations delivered by a set of special-purpose input systems that are now known as modules.

    If, therefore, we are to start with anything like Turing machines as models in cognitive psychology, we must think of them as embedded in a matrix of subsidiary systems which affect their computations in ways that are responsive to the flow of environmental events. The function of these subsidiary systems is to provide the central machine with information about the world. (Fodor, 1983, p. 39)

    According to Fodor (1983), a module is a neural substrate that is specialized for solving a particular information processing problem. It takes input from transducers, preprocesses this input in a particular way (e.g., computing three-dimensional structure from transduced motion signals [Hildreth, 1983; Ullman, 1979]), and passes the result of this preprocessing on to central processes. Because modules are specialized processors, they are domain specific. Because the task of modules is to inform central processing about the dynamic world, modules operate in a fast, mandatory fashion. In order for modules to be fast, domain-specific, and mandatory devices, they will be “wired in,” meaning that a module will be associated with fixed neural architecture. A further consequence of this is that a module will exhibit characteristic breakdown patterns when its specialized neural circuitry fails. All of these properties entail that a module will exhibit informational encapsulation: it will be unaffected by other models or by higher-level results of isotropic processes. In other words, modules are cognitively impenetrable (Pylyshyn, 1984). Clearly any function that can be shown to be modular in Fodor’s sense must be a component of the architecture.

    Fodor (1983) argued that modules should exist for all perceptual modalities, and that there should also be modular processing for language. There is a great deal of evidence in support of this position.

    For example, consider visual perception. Evidence from anatomy, physiology, and clinical neuroscience has led many researchers to suggest that there exist two distinct pathways in the human visual system (Livingstone& Hubel, 1988; Maunsell &Newsome, 1987; Ungerleider &Mishkin, 1982). One is specialised for processing visual form, i.e., detecting an object’s appearance: the “what pathway.” The other is specialised for processing visual motion, i.e., detecting an object’s changing location: the “where pathway.” This evidence suggests that object appearance and object motion are processed by distinct modules. Furthermore, these modules are likely hierarchical, comprising systems of smaller modules. More than 30 distinct visual processing modules, each responsible for processing a very specific kind of information, have been identified (van Essen, Anderson, & Felleman, 1992).

    A similar case can be made for the modularity of language. Indeed, the first biological evidence for the localization of brain function was Paul Broca’s presentation of the aphasic patient Tan’s brain to the Paris Société d’Anthropologie in 1861 (Gross, 1998). This patient had profound agrammatism; his brain exhibited clear abnormalities in a region of the frontal lobe now known as Broca’s area. The Chomskyan tradition in linguistics has long argued for the distinct biological existence of a language faculty (Chomsky, 1957, 1965, 1966). The hierarchical nature of this faculty—the notion that it is a system of independent submodules— has been a fruitful avenue of research (Garfield, 1987); the biological nature of this system, and theories about how it evolved, are receiving considerable contemporary attention (Fitch, Hauser, & Chomsky, 2005; Hauser, Chomsky, & Fitch, 2002). Current accounts of neural processing of auditory signals suggest that there are two pathways analogous to the what-where streams in vision, although the distinction between the two is more complex because both are sensitive to speech (Rauschecker & Scott, 2009).

    From both Fodor’s (1983) definition of modularity and the vision and language examples briefly mentioned above, it is clear that neuroscience is a key source of evidence about modularity. “The intimate association of modular systems with neural hardwiring is pretty much what you would expect given the assumption that the key to modularity is informational encapsulation” (p. 98). This is why modularity is an important complement to architectural equivalence: it is supported by seeking data from cognitive neuroscience that complements the cognitive penetrability criterion.

    The relation between modular processing and evidence from cognitive neuroscience leads us to a controversy that has arisen from Fodor’s (1983) version of modularity. We have listed a number of properties that Fodor argues are true of modules. However, Fodor also argues that these same properties cannot be true of central or isotropic processing. Isotropic processes are not informationally encapsulated, domain specific, fast, mandatory, associated with fixed neural architecture, or cognitively impenetrable. Fodor proceeds to conclude that because isotropic processes do not have these properties, cognitive science will not be able to explain them.

    I should like to propose a generalization; one which I fondly hope will someday come to be known as ‘Fodor’s First Law of the Nonexistence of Cognitive Science.’ It goes like this: the more global (e.g., the more isotropic) a cognitive process is, the less anybody understands it. (Fodor, 1983, p. 107)

    Fodor’s (1983) position that explanations of isotropic processes are impossible poses a strong challenge to a different field of study, called evolutionary psychology (Barkow, Cosmides, & Tooby, 1992), which is controversial in its own right (Stanovich, 2004). Evolutionary psychology attempts to explain how psychological processes arose via evolution. This requires the assumption that these processes provide some survival advantage and are associated with a biological substrate, so that they are subject to natural selection. However, many of the processes of particular interest to evolutionary psychologists involve reasoning, and so would be classified by Fodor as being isotropic. If they are isotropic, and if Fodor’s first law of the nonexistence of cognitive science is true, then evolutionary psychology is not possible.

    Evolutionary psychologists have responded to this situation by proposing the massive modularity hypothesis (Carruthers, 2006; Pinker, 1994, 1997), an alternative to Fodor (1983). According to the massive modularity hypothesis, most cognitive processes—including high-level reasoning—are modular. For instance, Pinker (1994, p. 420) has proposed that modular processing underlies intuitive mechanics, intuitive biology, intuitive psychology, and the self-concept. The mind is “a collection of instincts adapted for solving evolutionarily significant problems—the mind as a Swiss Army knife” (p. 420). The massive modularity hypothesis proposes to eliminate isotropic processing from cognition, spawning modern discussions about how modules should be defined and about what kinds of processing are modular or not (Barrett & Kurzban, 2006; Bennett, 1990; Fodor, 2000; Samuels, 1998).

    The modern debate about massive modularity indicates that the concept of module is firmly entrenched in cognitive science. The issue in the debate is not the existence of modularity, but is rather modularity’s extent. With this in mind, let us return to the methodological issue at hand, investigating the nature of the architecture. To briefly introduce the types of evidence that can be employed to support claims about modularity, let us consider another topic made controversial by proponents of massive modularity: the modularity of musical cognition.

    As we have seen, massive modularity theorists see a pervasive degree of specialization and localization in the cognitive architecture. However, one content area that these theorists have resisted to classify as modular is musical cognition. One reason for this is that evolutionary psychologists are hard pressed to explain how music benefits survival. “As far as biological cause and effect are concerned, music is useless. It shows no signs of design for attaining a goal such as long life, grandchildren, or accurate perception and prediction of the world” (Pinker, 1997, p. 528). As a result, musical processing is instead portrayed as a tangential, nonmodular function that is inconsequentially related to other modular processes. “Music is auditory cheesecake, an exquisite confection crafted to tickle the sensitive spots of at least six of our mental faculties” (p. 534).

    Not surprisingly, researchers interested in studying music have reacted strongly against this position. There is currently a growing literature that provides support for the notion that musical processing—in particular the perception of rhythm and of tonal profile—is indeed modular (Alossa & Castelli, 2009; Peretz, 2009; Peretz & Coltheart, 2003; Peretz & Hyde, 2003; Peretz & Zatorre, 2003, 2005). The types of evidence reported in this literature are good examples of the ways in which cognitive neuroscience can defend claims about modularity.

    One class of evidence concerns dissociations that are observed in patients who have had some type of brain injury. In a dissociation, an injury to one region of the brain disrupts one kind of processing but leaves another unaffected, suggesting that the two kinds of processing are separate and are associated with different brain areas. Those who do not believe in the modularity of music tend to see music as being strongly related to language. However, musical processing and language processing have been shown to be dissociated. Vascular damage to the left hemisphere of the Russian composer Shebalin produced severe language deficits but did not affect his ability to continue composing some of his best works (Luria, Tsvetkova, & Futer, 1965). Reciprocal evidence indicates that there is in fact a double dissociation between language and music: bilateral damage to the brain of another patient produced severe problems in music memory and perception but did not affect her language (Peretz et al., 1994).

    Another class of evidence is to seek dissociations involving music that are related to congenital brain disorders. Musical savants demonstrate such a dissociation: they exhibit low general intelligence but at the same time demonstrate exceptional musical abilities (Miller, 1989; Pring, Woolf, & Tadic, 2008). Again, the dissociation is double. Approximately 4 percent of the population is tone deaf, suffering from what is called congenital amusia (Ayotte, Peretz, & Hyde, 2002; Peretz et al., 2002). Congenital amusics are musically impaired, but they are of normal intelligence and have normal language abilities. For instance, they have normal spatial abilities (Tillmann et al., 2010), and while they have short-term memory problems for musical stimuli, they have normal short-term memory for verbal materials (Tillmann, Schulze, & Foxton, 2009). Finally, there is evidence that congenital amusia is genetically inherited, which would be a plausible consequence of the modularity of musical processing (Peretz, Cummings, & Dube, 2007).

    A third class of evidence that cognitive neuroscience can provide about modularity comes from a variety of techniques that noninvasively measure regional brain activity as information processing occurs (Cabeza & Kingstone, 2006; Gazzaniga, 2000). Brain imaging data can be used to seek dissociations and attempt to localize function. For instance, by seeing which regions of the brain are active during musical processing but not active when a nonmusical control task is performed, a researcher can attempt to associate musical functions with particular areas of the brain.

    Brain imaging techniques have been employed by cognitive neuroscientists interested in studying musical processing (Peretz & Zatorre, 2003). Surprisingly, given the other extensive evidence concerning the dissociation of music, this kind of evidence has not provided as compelling a case for the localization of musical processing in the human brain (Warren, 2008). Instead, it appears to reveal that musical processing invokes activity in many different areas throughout the brain (Schuppert et al., 2000). “The evidence of brain imaging studies has demonstrated that music shares basic brain circuitry with other types of complex sound, and no single brain area can be regarded as exclusively dedicated to music” (Warren, 2008, p. 34). This is perhaps to be expected, under the assumption that “musical cognition” is itself a fairly broad notion, and that it is likely accomplished by a variety of subprocesses, many of which are plausibly modular. Advances in imaging studies of musical cognition may require considering finer distinctions between musical and nonmusical processing, such as studying the areas of the brain involved with singing versus those involved with speech (Peretz, 2009).

    Disparities between behavioural evidence concerning dissociations and evidence from brain imaging studies do not necessarily bring the issue of modularity into question. These disparities might simply reveal the complicated relationship between the functional and the implementational nature of an architectural component. For instance, imagine that the cognitive architecture is indeed a production system. An individual production, functionally speaking, is ultra-modular. However, it is possible to create systems in which the modular functions of different productions do not map onto localized physical components, but are instead defined as a constellation of physical properties distributed over many components (Dawson et al., 2000). We consider this issue in a later chapter where the relationship between production systems and connectionist networks is investigated in more detail.

    Nevertheless, the importance of using evidence from neuroscience to support claims about modularity cannot be understated. In the absence of such evidence, arguments that some function is modular can be easily undermined.

    For instance, Gallistel (1990) has argued that the processing of geometric cues by animals facing the reorientation task is modular in Fodor’s (1983) sense. This is because the processing of geometric cues is mandatory (as evidenced by the pervasiveness of rotational error) and not influenced by “information about surfaces other than their relative positions” (Gallistel, 1990, p. 208). However, a variety of theories that are explicitly nonmodular are capable of generating appropriate rotational error in a variety of conditions (Dawson, Dupuis, & Wilson, 2010; Dawson et al., 2010; Miller, 2009; Miller & Shettleworth, 2007, 2008; Nolfi, 2002). As a result, the modularity of geometric cue processing is being seriously re-evaluated (Cheng, 2008).

    In summary, many researchers agree that the architecture of cognition is modular. A variety of different kinds of evidence can be marshaled to support the claim that some function is modular and therefore part of the architecture. This evidence is different from, and can complement, evidence about cognitive penetrability. Establishing the nature of the architecture is nonetheless challenging and requires combining varieties of evidence from behavioural and cognitive neuroscientific studies.


    This page titled 3.15: Modularity of Mind is shared under a CC BY-NC-ND license and was authored, remixed, and/or curated by Michael R. W. Dawson (Athabasca University Press) .

    • Was this article helpful?