Reading a modularised process?

Phonological processing, an important prerequisite for reading acquisition is viewed as a module, the selective damage of which is thought to lead to

dyslexia (see chapter 1). Reading is an a-theoretical term under which a multitude of psychological component processes are subsumed. These processes range from phonology and word recognition at a pre-lexical stage to high-level comprehension processes involving inference making and metacognition at post- lexical stages. Gough and Tunmer (1986) put forward a simple model of reading, comprising two component processes: word recognition and comprehension. Reading ability is thereby the product of word recognition and comprehension skills; deficits can occur in one subcomponent independently of the other. Conners and Olson (1990) suggest a very similar two-component model, where word recognition is composed of phonological and orthographic skills, and reading comprehension is best determined by listening comprehension, which in turn is dependent on general verbal ability and verbatim memory.

Both these models are useful as they split reading into processes that could be viewed as modular (or modularised), versus those that are a correlate of central processes or intelligence. Reading abilities appear to involve a modularised process on the one hand, and central processes on the other. Reading can be viewed as a combination of at least two main component processes: word recognition and comprehension. Relevant literature will be reviewed in the next section to show that word recognition can be viewed as a modularised process, while reading comprehension is best viewed as a central process.

2.4.1 Interaction of central and modularised processes

While phonological processing has been suggested to constitute a module in the Fodorian sense (Anderson, 1992; Stanovich, 1990; Shankweiler & Crain, 1986), word recognition can be seen as an acquired module or a module that comes into existence through a process of modularisation (e.g. Perfetti, 1992; Stanovich et al., 1985; Stanovich & West, 1983). In numerous studies it has been established that the effects of background knowledge and contextual information decline as word recognition efficiency increases (e.g. Pring & Snowling, 1986; Stanovich, 1986; 1980; Perfetti, 1985; Briggs, Austin & Underwood, 1984; Perfetti & Roth, 1981; Perfetti, Finger & Hogabaum, 1978). Word recognition thereby becomes automatic and is beyond the influence of strategic control, thus becoming gradually informationally encapsulated.

Perfetti et al. (1978), for example, varied the predictability of words by manipulating the set size from which they were drawn (e.g. the months in the calendar are a closed, and relatively small set; first names are a large, open set). They found that skilled third graders were nearly unaffected by set size, while less skilled readers were substantially affected, naming words from a large, open set especially slowly. Perfetti and Roth (1981) found that skilled readers, asked to name words that are counterpredicted in a given context, were hardly affected relative to low predictability contexts. Less skilled readers, however, were dramatically affected, in the way one might expect for a process that was slow and unreliable enough to allow specific word predictions to be generated and then disconfirmed. Perfetti and Roth (1981) also found that when words were visually degraded to slow down word recognition times, skilled readers showed context facilitation effects identical to those shown by less skilled readers on un­ degraded words. Thus simply slowing down a reader’s word recognition rate increases the facilitating effect. In skilled readers word recognition processes function sufficiently fast so that slower acting context effects cannot bear upon these. This reflects the relationship between a rapidly executing process (i.e. module) and the ability of some outside source of information to affect it (i.e. central processes).

Pring and Snowling (1986) examined the extent to which 8 and 10 year old children use context to facilitate decoding of novel words (i.e. nonwords). They were presented with word-nonword pairs, where the non words were always pseudohomophones. The target (i.e. pseudohomophone) and the preceding word were semantically related (e.g. doctor-nirse), unrelated (e.g. black-nirse), or neutral (e.g. xxx-nirse). The authors found that the 8 year old children showed a significant semantic facilitation effect, while this effect was not significant for the group of 10 year old children. These findings can be taken as evidence for the progressive encapsulation or modularisation of word recognition in the development of reading.

Stanovich and Cunningham (1991) note the two advantages of modularity; the “veridicality that results from the organism’s ability to code the features of the environment without distortions” (p. 14), and “the speed of encapsulated systems when the specificity and efficiency of stimulus analysing mechanisms is great relative to the diagnosticity of the background information”

(p. 15). They argue further that “the greater diagnosticity of the external stimulus in reading, as opposed to listening, puts a greater premium on an input system that can deliver a full representation of the stimulus to higher-level cognitive systems” (p. 16). This refers to the relatively low probability of predicting words from the context in which they appear. A reader’s probability of predicting the next word is between .25 and .35 (Stanovich & Cunningham, 1991). Consequently top-down processes or central processes are not very helpful in skilled word recognition, especially as they are slow and take up resources that are limited and required for higher-level comprehension processes. Context-free word recognition thus characterises skilled reading.

In another study, Pring and Snowling (1989) used a modified Stroop task, in which 8 and 10 year olds had to name the colour of the ink in which target words were presented. The target words were either preceded by semantically related or unrelated primes. The measure of ink colour naming time is sensitive to priming effects, in that the primes activate responses that compete with the required responses in the colour naming task, slowing these down considerably. Pring and Snowling (1989) found the older, better readers were influenced more by a preceding context that was semantically related than the younger less skilled readers. On a first viewing, these results seem to contradict those suggesting that context effects decrease with increasing reading skill. However, it is important to differentiate between pre-lexical stages of autonomous word recognition and post-lexical stages of editing and correcting of representations. The former correspond to automatic, modular processes that proceed without consultation; the latter to attentional or conscious processes that use central resources. Only the former’s reliance on context is thought to decrease with age and better word recognition abilities. In the colour naming study the children had to focus on the colour naming response, and so had to allocate their attention and central resources to this task. The influence of the words, so the argument goes, may be an automatic one, which is mediated through links between semantic and word recognition systems. It can be seen from this that it is not clear what exactly belongs in the domain of a “reading module”. Once the mapping between phonology and orthography has become established, decoding in an automatic, modular fashion should be possible. However, there may also be associated semantic representations. A network of phonological-orthographic and semantic

representations may constitute the domain of a ‘reading module’ most accurately, and cannot be penetrated by top-down processes, such as expectancies and inferences.

Nation and Snowling (in press) investigated the use of contextual cues by dyslexic readers, poor comprehenders and normal readers. Subjects were presented with printed words in two conditions: the words were either presented in isolation or following a spoken context. Accuracy and speed of responses were measured. Nation and Snowling (in press) found that all children’s reading was significantly better in the context condition. This context effect was greatest for the dyslexic group, and smallest for the group of poor comprehenders. It thus seems that dyslexic children use their central processing abilities in word recognition to a greater degree than normal children, or children who are poor comprehenders. The finding that all groups’ accuracy and speed was improved by contextual cues again can be explained in terms of progressive modularisation in development (a process impaired in dyslexia), or by the fact that modular and central processes may be confounded in the behavioural measures employed.

Stanovich and West (1983) explored the influence of context on the latencies to produce target words that are presented as completion of sentences. They compared recognition times for words that were contextually congruous, incongruous and neutral. They found that words in the congruous contexts were named faster relative to the neutral condition, when the target followed the context immediately. Latencies in the incongruous condition decreased only when a delay was introduced between the context and the target. Stanovich and West (1983) argued that facilitation is due to an automatic spreading mechanism within the lexicon, uninfluenced by expectancies. On the other hand, inhibition effects are due to controlled attentional processes, reflecting the activity of the comparatively slow central processes that set in when the modular process has been completed, operating on the modules’ output representations. This study thus presents a successful attempt at disentangling modular and central processes empirically. Given that semantic factors act fast and mandatorily suggests again that at least some level of semantic representation is part of the domain of what one might call the ‘reading module’.

In conclusion, phonological processes can be viewed as a module in the Fodorian sense, while word recognition may be an acquired module or a

modularised component of the cognitive architecture. A culturally acquired skill such as reading cannot be fitted into Fodor’s nativist account of modularity but, once fiilly developed may still share the essential characteristics of a module.