General and specific cognitive processes: the modularity thesis
2.1 The modularity thesis
This thesis aims to investigate modular and central processes in developmental disorders. In this chapter Fodor’s (1983) work on modularity and central processing, and Karmiloff-Smith’s (1992) related concept of modularisation is reviewed. Fodor’s (1983) modularity concept is used to explore several candidate processes: phonology as a module and decoding as a modularised process, and intelligence and reading comprehension as central processes.
In his model Fodor (1983) conceptualises the mind as consisting of a number of input systems or modules. These systems, which may be considered as extensions of the sensory systems, provide input to higher level processes that Fodor terms central processes. To qualify as a Fodorian module, a psychological process must fulfill nine criteria: domain-specificity, mandatory operation, impenetrable (i.e. only final output is accessible to central processes), fast, informationally encapsulated, delivering shallow outputs, fixed neural architecture, and exhibiting a specific pattern of breakdown and ontogeny (pace and sequencing).
Domain-specificity refers to the specificity of the range of stimuli that a module can operate upon. A module is an expert system with a circumscribed area of processing that corresponds to a psychological domain. Fodor gives the example of the perceptual systems that effect the phonetic analysis of speech. If segments of an utterance are isolated then the hearer is unable to identify these isolated segments as part of language; instead these are perceived as “whistles” or “glides” (Liberman et. al., 1967), Fodor thus concludes that the input analyser or module dedicated to this process is capable only of analysing a very specific area of auditory perception (i.e. continuous speech).
The second criterion is that the operation of modules is mandatory. This means that the processing of a module is automatic and is obligatorily applied; in other words this processing cannot be stopped intentionally and is as such inflexible and insensitive to demands made by the conscious parts of the mind. Fodor supports this notion by observing that “you can’t help hearing an utterance as an utterance, or seeing a visual array as consisting of objects distributed in three-dimensional space”, (Fodor, 1983, p.52).
Thirdly, central processes have only limited access to the mental representations that modules compute. Modules receive their input from sensory systems that Fodor calls “transducers” and transform this input into representations that central processes can compute. This occurs via several so called “interlevels” of representations. Central processes have access only to the final representations of the modules; interlevel representations are inaccessible. Again Fodor gives an intuitively appealing example supporting this notion: “Not only must you hear an utterance of a sentence as such, but you can hear it only that way” (p.56).
Fourthly, input systems are fast processors. This seems to follow in part from mandatory and automatic processing; no time is devoted to decision making. In addition, modules are on-line processors which work immediately and are not driven from representations in memory. Consequently, they have to process at least as fast as the rate at which the more peripheral sensory systems provide input to them. Fodor points out that identification of sentences and visual arrays is among the fastest of our psychological processes, and gives the example of shadowing of speech, which can be accomplished with a quarter second latency (Marslen-Wilson, 1973). Within this quarter second, perceptual analysis and integration of the speech, repetition and comprehension of the utterance is accomplished.
The fifth criterion, informational encapsulation, captures the essence of Fodor’s modularity thesis. Modules are closed systems in that they are unaffected by feedback from higher processes. This is a very strong claim that at first viewing stands in opposition to many recent findings in psychology, especially from the area of psychology termed ‘New Look’ psychology (Bruner, 1973). Findings such as the phoneme restoration effect (Warren, 1970), the effects of context on word recognition and sentence recognition, and in the visual domain.
the maintenance of perception in case of scotoma on the retina, all seem to suggest that information feedback upon input operation is present. This in turn, led to the proposal of top-down models, where the encoding of stimuli is affected by the subject’s beliefs and expectations. Fodor argues that all this evidence does not, however, show that this feedback penetrates down to intramodular levels, but only that it may be operative upon the final representations delivered by the modules. Fodor puts forward some caveats in regard to interpreting data that presumably shows penetrability of input systems. He argues that in order to disprove his thesis, one would need to show that the locus of top-down effect is internal to the module, and that simply showing penetrability of processes that may be similar to modular processes, is not in itself evidence for the penetrability of modules. Further, he warns that informational encapsulation is not equivalent to the suggestion that no top-down processes operate within modules. In fact, processes internal to a module may mimic effects of cognitive penetration. To make his point clear, he gives the example of context facilitation in sentence recognition. In a sentence context about a microphone and a spy, the word bug facilitates recognition of the word microphone, but to an equal extent facilitates activation of the word insect (Swinney 1979). These effects, he argues are better explained in terms of lexical network associations, in which recognition thresholds change with spreading activation (Morton, 1969; Collins & Loftus, 1975; Meyer & Schyaneveldt, 1971), rather than in terms of explicit knowledge by the reader. The lexical network is thereby part of the language module, and top-down processing within this domain is possible. Consequently, the encapsulation leads to an advantage in speed of processing, as decision time is not needed, and the confirmation possibilities of perceptual stimuli are restricted. The modules “buy speed at the price of unintelligence”, (p.80).
Informational encapsulation hinges on the next criterion that modules have ‘shallow’ outputs. Shallow outputs result as a consequence of the constrained domain of information that a module processes, and can be viewed as the opposite of global meaning or meaning in context. Informational encapsulation thus has advantages when exact stimulus analysis without top- down processing is required. On the other hand, for processing of stimuli that require integration and evaluation in context, informational encapsulation would be a disadvantage. The question is thus, where is the line between perception and
cognition, or between modular and central processing to be drawn? Fodor makes a few speculative claims, concentrating on the deepest level that modules may have as their outputs. He offers two examples of processes that are relatively non-shallow, but may still be sufficiently shallow to fulfill the criteria of modular functioning. One is the definitional content of words in sentences, and another the perception of basic-level categories (Brown, 1958; Rosch, 1976). Fodor points out that a conclusion, on what the boundaries of a module may constitute, will be reached through empirical research only, and will thus be answered in the future.
The seventh criterion refers to the fixed neural architecture of modules. Fodor (1983) argues that an informationally encapsulated system is best represented in the brain as a discrete area with fixed neural architecture. The nature of information processing within a module is complex, but stable over time so that specialisation in terms of neurological structure is of benefit here. The specific neurological structures in the brain associated with perceptual and with language systems support this notion of fixed neural architecture. As opposed to this, non-modular processes would be served by uniform brain structures, that initially may be characterised by equipotentiality and plasticity.
Criterion eight follows from the previous one: If modules have fixed neural architecture, then they will exhibit characteristic and specific breakdown patterns. Fodor does not elaborate this point in any great detail, other than saying that well-defined pathological syndromes exist in the perceptual as well as the language processing domains (i.e. agnosias and aphasias). However, he argues that specific breakdowns could just as well result from the damage of a merely functionally distinct process, such as memory or attention. Thus inferring the workings of a module on the observation of a specific breakdown of functioning alone may be unwarranted.
Finally, the last criterion: ontogeny of input systems exhibits a characteristic pace and sequencing. Again, this idea is linked to the fixed neural architecture criterion. Fodor puts this criterion forward as the most tentative, saying that so far no facts available contradict this claim, and the claim itself as even more vague: “a great deal of the developmental course of the input systems is endogeneously determined”. This follows from Fodor’s ideas of the evolution of modularity and its genetic determination.
In summary, a module is not ‘intelligent’, as no conscious thought is involved. No planning of action is possible. As a long as a module receives input it will compute, regardless of whether this information is required at a particular moment, and whether or not attention will be paid to it. Consequently, modules are highly specialised and able to process very complex information at high speeds.
2.1.2 What is central processing?
Central processes do not share any of the characteristics of modules. They are provided only with the final computational output of the modules (i.e. representations). These representations from the various modules need to be evaluated in conjunction and possibly corrected in the light of background knowledge. Fodor (1983) calls this process of constructing corrected representations, “the fixation of perceptual belief’, (p. 102), and “the computation of best hypotheses about what the world is like”, (p. 104). Fodor (1983) then makes an analogy to scientific hypothesis confirmation, in that the modules provide clues about the not directly observable world (i.e. empirical data) and the central systems use this data for hypothesis testing (i.e. the scientist’s activity). He further argues that science is both isotropic and Quinean. Isotropic means that information that may be relevant to the confirmation of a hypothesis may be drawn from anywhere in the knowledge base; Quinean refers to the fact that any confirmation of a hypothesis is dependent on the properties of the entire belief system. From this it becomes clear that central processes are global and unencapsulated, requiring the intentional or conscious evaluation of information. Given these characteristics, Fodor (1983) argues that one would expect unstable, instantaneous connectivity that changes from moment to moment rather than discrete anatomical structure as in the case of the modular neural architecture. However, the characterisation of central processes remains vague, and Fodor justifies this with the nature of science as a tool that parses the observable into its constituent parts. As the definition of central processes is the opposite (i.e. characterised by holistic functioning), Fodor concludes his monograph with his first “Law of the Nonexistence of Cognitive Science”: “The more global a cognitive process is the less anybody understands it“, (p. 107). Central processes.
in Fodor’s definition, are defined at an abstract level, and to some extent merely as default to being non-modular.