Modularity and Language

Is language a separate mental faculty? Or is it just part of a monolithic general-purpose cognitive system? The idea that the human mind is composed of distinct faculties was hotly debated in the nineteenth century, and the debate continues today. Genetic disorders such as Williams syndrome show that individuals who cannot count to three or solve simple spatial tasks nevertheless develop remarkable language skills that resemble those of a fully fluent and proficient second-language learner (Bellugi et al. 1993; Karmiloff-Smith et al. 1997). The striking disparity in the levels of attainment of Williams syndrome individuals in different cognitive domains clearly argues for differentiated mental capacities.

Studies of the brain lead to the same conclusion. The left hemisphere of the brain is the language-dominant hemisphere for right-handed individuals. In 1861, Paul BROCA identified the third frontal gyrus of the language-dominant hemisphere as an important language area. Performing autopsies on brain-damaged individuals with expressive difficulties characterized by slow, effortful "telegraphic" speech, he found that their lesions involved the third frontal gyrus, now known as "Broca's area." Using modern neuroimaging techniques, Smith and Jonides (1997) have implicated Broca's area specifically in the rehearsal of material in verbal WORKING MEMORY in normal adults, showing an increase in activation with increasing memory load. Spatial tasks requiring active maintenance of spatial information in working memory do not activate Broca's area in the left hemisphere but rather the premotor cortex of the right hemisphere. Though the overall picture of language representation in the brain is far from clear, the debates today mostly concern, not whether areas specialized for language (or object recognition or spatial relations) exist, but how these areas are distributed in the brain and organized.

In studies of normal adult language processing, modularity is discussed in broad terms where the questions concern the separability of grammatical processing from general cognitive processing and in narrow terms where the questions concern the isolability of distinct components of the grammar. One central question is whether a syntactic parser exists that is concerned only with the construction of syntactic structure in production or the identification of syntactic structure in comprehension. In studies of sentence production, Bock (1989) presents evidence for purely syntactic priming not dependent on semantic content or the particular words in a sentence. In other words, having just produced a sentence with a particular syntactic structure, speakers tend to use that same structure again even under circumstances where there is no semantic relation between the two sentences. This result is expected if syntax is a specialized component of a modular language processor in the narrow sense. Similarly in comprehension, Clifton (1993) has shown that a phrase following an optionally transitive verb is first analyzed as an object of that verb even if semantic properties of the clause dictate that ultimately it must be the subject of a following clause. Both studies provide evidence for the existence of autonomous syntactic structures that participate in language processing.

Laying out a very specific modularity thesis, Fodor (1983) hypothesizes that perceptual, or "input," systems share several important characteristics. They apply only to a limited domain, visual inputs, for example. Only domain-specific information (e.g., visual information) is applied in the input system. The operation of the input system is fast and reflexive (automatic and mandatory). There is limited access to the intermediate representations computed -- essentially only the final output of the system is available to other systems. Each input system is biologically determined in the sense that its development exhibits a characteristic pace and sequence, it has an associated neural basis, and when that neural basis is damaged, characteristic deficits result. Fodor claims that language is an input system -- it exhibits all of the properties of a perceptual system.

To argue that the language system, or more accurately the grammatical system, is domain-specific, Fodor notes that perceivers integrate auditory and visual information about a speech input, perceiving an "average" when the two conflict. Presented with a videotaped speaker forming a consonant in the back of the mouth ("ga") and a synchronized auditory input "ba," perceivers will integrate the forward (labial) articulation of the "ba" with the evidence about the place of articulation of "ga" from the video and perceive "da" -- a sound produced farther back in the mouth than labials such as "ba" but farther forward than sounds such as "ga." This is known as the "McGurk effect" (McGurk and MacDonald 1976). The point about the McGurk effect is that it is not simply a guess on the part of confused perceivers about how they can resolve conflicting perceptual inputs. It is an actual perceptual illusion, as expected if SPEECH PERCEPTION is part of an input system specialized for speech inputs.

The most controversial aspect of Fodor's thesis is the claim that language processing is "informationally encapsulated," that only domain-specific information is consulted within a module. The question is whether the massive effects of nonlinguistic world knowledge actually occur after an initial hypothesis has already been identified within the language module on the basis of purely linguistic knowledge or whether world knowledge can direct the grammatical processing of the input. Word recognition is one area where this debate has been played out.

Word recognition studies demonstrate that both meanings of an ambiguous word are activated (Swinney 1979), at least when the word occurs in a semantically neutral sentence. In a semantically biased sentence favoring the more frequent meaning of the ambiguous word, only the dominant (frequent) meaning of the word is activated, suggesting that perhaps word recognition is not informationally encapsulated. On the other hand, in a sentence biased toward the less frequent meaning of the ambiguous word, both the contextually appropriate and the contextually inappropriate meaning of the ambiguous words are activated, as we would expect if word recognition were informationally encapsulated. Proponents of modularity take heart in this latter finding and explain the former finding in terms of the frequent meaning being activated and accepted so quickly that it can inhibit the activation of the less frequent meaning (Rayner and Frazier 1989). Opponents of modularity focus on the former finding and note that it indicates that context can influence word recognition under at least some circumstances (Duffy, Morris, and Rayner 1988, for example).

Ultimately, the survival of Fodor's modularity thesis may depend on its explanatory value. Precisely because a module operates mandatorily and consults only restricted information (identifiable in advance of any particular input), the identification, access, or computation of information can be fast. The grammatical processor's job is a structured and limited one.

If the grammar or grammatical subsystems act as modules, it also becomes less surprising that grammars have the eccentric properties that they do, relying on strict module-internal notions of prominence such as "c-command" (Reinhart 1983), rather than on generally available notions based on, say, precedence, loudness, or the importance of the information conveyed. For many linguists, it is the reappearance within and across languages of the same peculiar notion of prominence or locality that most convincingly argues that grammars form, not loose associations of "biases" or co-occurrence probabilities, but specialized systems or modules.

See also

Additional links

-- Lyn Frazier


Bellugi, U., S. Marks, A. Bihrle, and H. Sabo. (1993). Dissociation between language and cognitive function in Williams syndrome. In D. Bishop and K. Mogford, Eds., Language Development in Exceptional Circumstances. Mawah, NJ: Erlbaum.

Bock, K. (1989). Closed class immanence in sentence production. Cognition 31:163-186.

Clifton, C. (1993). Thematic roles in sentence parsing. Canadian Journal of Psychology 47:222-246.

Duffy, S. A., R. K. Morris, and K. Rayner. (1988). Lexical ambiguity and fixation times in reading. Journal of Memory and Language 27:429-446.

Fodor, J. A. (1983). Modularity of Mind. Cambridge, MA: MIT Press.

Karmiloff-Smith, A., J. Grant, I. Berthoud, M. Davies, P. Howlin, and O. Udwin. (1997). Language and Williams syndrome: How intact is "intact"? Child Development 68(2):246-262.

McGurk, H., and J. MacDonald. (1976). Hearing lips and seeing voices. Nature 264:746-748.

Rayner, K., and L. Frazier. (1989). Selection mechanisms in reading lexically ambiguous words. Journal of Experimental Psychology: Learning, Memory and Cognition 15:779-790.

Reinhart, T. (1983). Anaphora and Semantic Interpretation. London: Croom Helm.

Smith, E. E., and J. Jonides. (1997). Working memory: A view from neuroimaging. Cognitive Psychology 33:5-42.

Swinney, D. (1979). Lexical access during sentence comprehension: (Re-)consideration of context effects. Journal of Verbal Learning and Verbal Behavior 18:645-659.