The Adaptable Mind: What Neuroplasticity and Neural Reuse Tell Us about Language and Cognition - John Zerilli 2021
Modular Notation
Modules Reconsidered: Whither Modularity?
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 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 (or even vision!) 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}
where M1, M2, etc., are modules. 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 N1, N2, etc., are neurons, so that a given module Ma will comprise a set of neurons
Ma : {Na, Nb, Nc, . . .}
An M-network is (or resembles) the structure that neuroscience variously terms a “module,” “column,” or “elementary processing unit,” and that 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 and 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}
where C1, C2, etc., are C-networks, 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 that 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 that C-networks serve to implement. But, as I have suggested several times already, the same scientists often 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 the 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 that is no longer tenable, pending further notice (Gold & Roskies 2008, p. 354).
To sum up, then, there are at least two networks of interest as far as the modularity of mind is concerned: the network of neurons that constitutes a node/M-network, and the network of nodes that 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) can be reduced 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.