Cognitive Neuropsychology and Developmental ...

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 1997, 50A (4), 899_ 923

Cognitive Neuropsychology and Developmental Disorders: Uncomfortable Bedfellows D.V.M. Bishop MRC Applied Psychology Unit, Cambridge, U.K.

Cognitive neuropsychology provides a theoretical framework and methods that can be of value in the study of developmental disorders, but the ``dissociation’’ logic at the centre of this approach is not well suited to the developmental context. This is illustrated with examples from speci c language impairment. Within the developing language system there is ample evidence for interaction between levels of representation, with modularity emerging in the course of development. This means that one typically is seeking to explain a complex pattern of associated impairments, rather than highly selective de cits. For instance, a selective impairment in auditory processing can have repercussions through the language system and may lead to distinctive syntactic de cits that are seen in written as well as spoken language. Changes in the nature of representations and in the relationships between components of a developing system mean that cross-sectional data at a single point in development may be misleading indicators of the primary de cit. Furthermore, traditional cognitive neuropsychology places a disproportionate emphasis on representational (competence) de cits, with processing (performance) de cits being relatively neglected. Methods for distinguishing these two kinds of impairment are discussed, as well as other approaches for elucidating the underlying nature of developmental disorders.

Cognitive Neuropsychology, Dissociated De® cits and Modularity John Marshall’s EPS 50th Anniversary lecture described how the Experimental Psychology Society provided a stimulating forum for interactions between cognitive psychologists and neuropsychologists in the 1960s and 1970s, leading to the owering of the discipline of cognitive neuropsychology. From the start, the aim of the enterprise was to develop and test models of the cognitive processes involved in particular mental operations by This paper is based on an invited presentation to the Experimental Psychology Society in Cambridge, July 1996, in celebration of the 50th anniversary of the founding of the Experimental Psychology Group, from which the Society subsequently evolved. Requests for reprints should be sent to D.V.M. Bishop, MRC Applied Psychology Unit, 15, Chaucer Road, Cambridge, CB2 2EF, U.K. Email: [email protected] This article is based in part on chapter 9 in D.V.M. Bishop, Uncommon understanding: development and disorders of language comprehension in children, published by Psychology Press, 1997.

Ó 1997 The Experimental Psychology Society

900

BISHOP

analysing patterns of de cit in brain-damaged individuals (see Coltheart (1987) and Shallice (1988) for reviews). The emphasis was largely on case studies that demonstrated highly selective patterns of cognitive de cit and could inform theory by demonstrating the relative independence of different cognitive processes. A great deal of evidence has accumulated to support the view that not only is language processing largely independent of other cognitive functions, but within the language system there are independent stages that may be selectively disrupted. For instance, one may see patients who lose the ability to understand spoken language while retaining the ability to repeat (e.g. Berndt, Basili, & Caramazza, 1987), or who are able to read content words but not function words (Gardner & Zurif, 1975); others may have particular dif culty in understanding sentences where word order is critical (Schwartz, Saffran, & Marin, 1980); in yet other cases, there may be loss of the ability to name or comprehend items of a given semantic class (e.g. Warrington & McCarthy, 1987). Evidence such as this has been used to develop information processing models of language comprehension as involving a sequence of repesentational stages, each largely independent of the other (e.g. Figure 1). The publication of Fodor’s (1983) in uential book, The Modularity of Mind, incorporated many of the ideas that had developed in cognitive neuropsychology by proposing that much of our cognitive processing is modular. According to Fodor, a module is a domain-speci c and informationally encapsulated brain system responsible for handling a particular type of mental representation, which has the properties listed in Table 1. This may be illustrated by considering how the brain handles speech perception. Domain speci city refers to the fact that the mechanisms involved in speech perception appear to be distinct from those used for other kinds of auditory analysis and operate only on acoustic signals that are taken to be utterances. Information encapsulation refers to the fact that modular processing cannot be in uenced by higher cognitive operations. This results in mandatory processing: one cannot easily ignore incoming speech—even if attention is switched away from interpreting utterances, one will immediately respond on hearing a salient signal such as one’s own name. Processing in a modular system is fast and automatic, in contrast to central processes concerned with long-term memory and reasoning, which are typically slow, optional, and general-purpose, able to combine information froma variety of sources. In Fodor’s scheme, processing of language is modular up to the stage where a propositional representation is generated. Later stages of understanding—that is, incorporating general and social knowledge to interpret the utterance— are seen as involving central processes and do not have the characteristics of modules. Fodor used evolutionary arguments to support the idea that a complex processor that rapidly performs a dedicated function (i.e. a module) is likely to be innately pre-programmed, and the nding that damage to a particular brain region can lead to a selective impairment in a particular stage of language processing has been taken as evidence for this.

FIG. 1. Information processing model showing the sequence of representations generated in comprehending the sentence ``the sh is on the table’’ . Pragmatic in uences on interpretation (handled by Fodor’s so-called ``central processes’ ’) are not shown, and processing is depicted as entirely bottom-up.

901

902

BISHOP TABLE 1 Fodor’s Characterization of Properties of a Module

domain speci city system constrained in terms of the range of information it can access mandatory processing no voluntary control over whether relevant input is processed limited central access to intermediate representations information is not available to conscious awareness. E.g., auditory characteristics of speech sounds, or precise syntactic form of an utterance are dif cult or impossible to report, even though the utterance containing these was understood speed complex information processing takes place remarkably quickly information encapsulation information from higher levels is not fed back to lower ones (e.g., no top down processing occurs) shallow output computes only a very limited range of representations xed neural architecture handled by a circumscribed and dedicated brain region characteristic breakdown patterns associated with selective de cits in one area of functioning that cannot be explained in terms of some general loss of capacity characteristic pace and sequencing in development developmental course of a modular function is highly dependent on maturation of endogenous systems, and insensitive to environmental in uences

The claim that modules are innate is important, because it implies that the logic of the cognitive neuropsychology approach can be applied to developmental as well as acquired disorders. If we have speci c brain regions that are already specialized for language processing at birth, then it seems reasonable to suppose that a child who is impaired in a modular process must have suffered damage to, or maldevelopment of, such a system. Enthusiasm for modularity has led to mounting interest from those studying developmental disorders, with a new discipline of developmental cognitive neuropsychology being spawned (Temple, 1997). The information processing models derived from the study of adults are seen as providing a useful framework for understanding the levels of representation that need to be investigated, and there has been growing awareness of developmental analogues of some acquired disorders, such as prosopagnosia (Young & Ellis, 1992), subtypes of reading impairment (Castles & Coltheart, 1993), and dyscalculia (Temple, 1992). In my own research, however, I have experienced mounting dissatisfaction with cognitive neuropsychology. In large part, the dif culties that I outline arise when applying this approach to a developing cognitive system. However, more general problems with the cognitive neuropsychology approach as a whole have also become apparent. My aim in this paper is not to recommend that we abandon cognitive neuropsychology or that we cease to make contact between developmental and acquired disorders. As I hope to demonstrate, I feel that a great deal can be learned by comparing de cits that are seen in children and adults, and methodologies and models developed in one context may be adapted for use in the other. My message, rather, is that we need to contrast as well as

DEVELOPMENTAL COGNITIVE NEUROPSYCHOLOGY

903

compare, to recognize what is special about the developmental as opposed to the acquired case. I hope that readers will nd this critique constructive, and that consideration of the issues that arise in a developmental context might feed back and stimulate new thinking among those whose primary concern is adult neuropsychology.

Speci® c Language Impairment in Children To illustrate my points, I shall draw predominantly on examples from the topic I know best—speci c language impairment (SLI), also known as ``developmental dysphasia’’ or ``developmental language disorder’’. This is de ned as a developmental disorder where the child’s language acquisition is, for no apparent reason, out of keeping with other aspects of development. Nonverbal IQ and hearing are within normal limits (see Bishop, 1994a, for an overview). Here we have a disorder that involves, by de nition, dissociation between cognitive domains that are normally associated (language and nonverbal ability), and so it might appear to be an excellent candidate for application of the kinds of models and techniques adopted so successfully by cognitive neuropsychologists in the study of acquired aphasia. And, to be sure, SLI, in common with other developmental disorders (speci c reading impairment, autistic disorder, developmental co-ordination disorder), does illustrate that we need more than a single dimension of IQ to account for variations in children’s cognitive development. The problems arise when we attempt to identify the primary underlying de cit by looking for fractionations within the language system. There is no shortage of candidate theories (see Bishop, 1992, 1997, for reviews). One in uential body of work proposes that the language dif culties are caused by an impairment in discriminating rapid or brief auditory stimuli (Tallal & Katz, 1989). According to this model, the breakdown occurs at a very early point in the information processing chain, somewhere between a representation of an acoustic waveform and it interpretation as a matrix of phonetic features (see Figure 1). Another theory attributes the dif culties to limitations in phonological short-term memory (Gathercole & Baddeley, 1990). On this view, a phonological representation is generated from spoken input, but it decays rapidly, and this affects the child’s ability to learn words and grammar. A third account maintains that specialized mechanisms for grammar acquisition are impaired in SLI (e.g. Crago & Gopnik, 1994), so that the child cannot generate adequate representations at the levels of phrase structure and/ or cannot map phrase structure onto thematic roles (see Figure 1). My case is that the traditional logic of cognitive neuropsychology is inadequate to discriminate these possibilities.

LIMITATIONS OF COGNITIVE NEUROPSYCHOLOGY I.

Problems for Developmental Applications 1. Focus on Dissociations Rather Than Associations

The emphasis that cognitive neuropsychology places on studying dissociations means that attention is focused on rare cases with unusual patterns of de cit. The justi cation is that a single case of dissociation is theoretically more informative than one thousand cases with multiple impairments, because it demonstrates a lack of logical dependence between

904

BISHOP

de cits. The fact that patients with striking patterns of dissociation are the exception rather than the rule is irrelevant, because the main aim of the enterprise is to build a model of normal cognition rather than to document the dif culties of neuropsychological patients in general. Some have gone so far as to argue that the study of single cases is the only valid means of proceeding, and that group studies should not be used if one’s aim is to develop a model of the architecture of the cognitive system (Caramazza, 1986). The problem with developmental disorders is that, although one may see fractionation along the lines of major domains of functioning (e.g. language, spatial perception, motor co-ordination, social cognition, numeracy) within any one domain, the typical observation is one of a complex pattern of associated impairments rather than the highly selective de cits that may be found in acquired disorders. And this should not surprise us, because, in the developing child, an impairment at an early stage of processing would affect all the processes downstream of that stage. This contrasts with the situation in acquired disorders in adults. Saffran, Marin, and Yeni-Komshian (1976), for instance, described a patient with ``word deafness’’, who was quite unable to do any tasks requiring perceptual analysis at the phonemic level. Although he could not understand spoken language, semantic and syntactic processes remained intact in his expressive speech. Furthermore, he could understand written language. In relation to the model shown in Figure 1, written input enabled him to generate a representation in the form of a sequence of words while bypassing the need for auditory and phonological analysis. Contrast this with the case of a child who, from the earliest stage of language acquisition, has a problem in decoding speech sounds. A disruption at this stage of processing would lead to impairment at all subsequent levels, because the systems that would normally be responsible for vocabulary learning and mastery of grammar will not receive adequate input. In such a case, we would expect to see a very different pro le of impairment in the child compared to the adult. Furthermore, we would expect the child to have dif culty in learning to read, because the necessary substrate of spoken language skills would not be acquired. To take another example, we know that in adults a brain lesion can severely impair phonological short-term memory, without any obvious effects on comprehension or speech production (Vallar & Baddeley, 1984). However, learning of new vocabulary is impaired (Baddeley, Papagno & Vallar, 1988). In an adult, whose vocabulary is already established, this may be just a minor inconvenience. However, in a young child who is still learning language, the consequences would be much more serious. In short, for the child who is still developing language, a selective impairment at an early stage of processing will have repercussions throughout all subsequent stages. The fact that children are especially likely to have associated de cits is not a coincidence—it is inevitable, given the interdependence of different stages of processing upon one another in the course of development. Single-case methodology is not helpful for studying associations, because we cannot establish which correlated impairments are just chance associations and which correspond to reliable patterns of co-occurrence. To address this issue, we need group studies. It cannot be emphasized enough that a focus on group studies does not preclude an interest in individual differences. The de cits in children with SLI are not just complex, they are also heterogenous, and so a critical question is how to interpret variation in language pro les. Variation from child to child could mean that we are in fact dealing with a group


905

of different disorders, in which case it is important to nd ways of cleanly distinguishing between them. However, we must remember that some variation may just be meaningless noise; tests are never perfect indicators of underlying cognitive processes and may be subject to in uences of attention, motivation, and other random uctuations (see Point 7). Furthermore, some variation in test performance might re ect individual differences in cognitive pro le similar to those seen in the normal population; after all, if we tested any group of children on a battery of language measures, we would expect to nd different pro les of strength and weakness from one to the next (cf. Goldberg, 1995). Clearly, there is no sense in reducing group data to a composite mean, when this is not representative of any individual in the group. However, the answer is not to restrict attention to the study of individual cases, but, rather, to identify reliable clusters of de cit; only by studying groups of individuals can we begin to disentangle what is systematic signal and what is noise from the complex patterns of impairment that are seen in a disorder such as SLI. 2. Focus on Bottom-up Processing

If we accept that, when dealing with a developmental disorder, the problem is not so much one of identifying a dissociation, as of trying to account for a cluster of associated impairments, then it might seem a reasonable strategy to search for the earliest stage of processing at which impairment could be seen. For instance, suppose a child has weak vocabulary, poor understanding of syntax, and dif culty in discriminating between speech sounds. If we accept Fodor’s proposition that processing of language input is handled by an informationally encapsulated modular system, where processing is strictly bottom-up (depicted in Figure 1), then we could conclude that the speech discrimination de cit re ects impairment to an earlier stage of processing than the vocabulary or syntactic dif culties and could hence be regarded as the primary de cit that in uences all subsequent stages of processing. However, this ignores the ample evidence for interaction between levels in language development. A more realistic model of part of the system is shown in Figure 2. Note that this includes top-down as well as bottom-up in uences. Consider, for instance, the path shown feeding back from syntax to lexicon. It is often assumed that children learn word meanings simply by recognizing familiar strings of phonemes and deducing the meaning from the environmental context. However, as Gleitman (1994) has cogently argued, although this strategy might be useful for learning concrete nouns, it is of little help in deducing meanings of other parts of speech, such as verbs, or abstract nouns. Furthermore, if the visual context were the major cue to learning meaning and structure, then congenitally blind children should have major dif culties in langauge acquisition, but, on the whole, they do not. Landau and Gleitman (1985) considered this issue in an analysis of the language of a blind child who learned to use verbs like ``look’’ and ``see’’ appropriately. They proposed that, once some grammatical knowledge is available, the child may perform a syntactic analysis on an input sentence containing an unfamiliar word and deduce its meaning from the syntactic frame (syntactic bootstrapping). For instance, if an adult describing a cartoon to a blind child says: ``Tom really walloped Jerry’’, the child can deduce that the word wallop refers to an action in which Tom was agent and Jerry was object. The grammatical elements associated with a verb provide information as to

906

BISHOP

FIG 2. A simpli ed model of early stages of language comprehension, indicating top-down as well as bottomup in uences.

whether it refers to self-caused actions (typically expressed by an intransitive verb—i.e. with a subject and no object, such as ``stand’’ or ``swim’’), an act that affects another’s state (typically a transitive verb with subject and direct object, such as ``kick’’ or ``tickle’’), or a propositional attitude (where the verb typically takes a sentence as a complement, such as ``think’’ or ``know’’). So if the child has to deduce what ``gorp’’ means, a different conclusion will be reached if the adult says ``John is gorping’’, ``John is gorping Max’’, or ``John gorps that Max is nice’’. It follows that, in development, syntax and vocabulary are not independent. The child who has poor syntactic skills will have reduced access to this important source of information about word meaning, and so vocabulary learning will be retarded. Next, consider the path back from lexicon to phonological processing. Evidence for this route comes from studies of phonological short-term memory. One might imagine that the task of repeating nonwords is a pure measure of phonological short-term memory, uncontaminated by vocabulary knowledge. However, we now know that the ability to remember novel strings of phonemes is in uenced by their wordlikeness (Gathercole, Willis, Emslie, & Baddeley, 1991). Thus, although theoretical interest initially focused on the question of how phonological STM in uences vocabulary development, we now recognize that a correlation may re ect in uences in the reverse direction: quite simply, the child with a large vocabulary has more opportunity to use knowledge of phonological structure in words to guide performance. These are just two examples of top-down in uences that have been shown to operate in the course of development. Overall, Fodor’s notion of a modular system that operates entirely in bottom-up mode might be a reasonable characterization of the stable state that is achieved in the adult once language is fully learned (although this is debatable—see Marslen-Wilson & Tyler, 1987); however, it provides an unrealistic model of processing in the child who is developing language, where there is ample evidence of top-down in uences on earlier stages of processing (see Bishop, 1997, for further examples). A similar


907

point has been made in the context of developmental dyslexia by Hulme and Snowling (1992), who concluded that `ìnteraction between systems is probably the norm in development, and it may only be after a very extensive period of development that the relative modularity or autonomy of different systems in the adult is achieved’’. They note the dif culties this leads to in trying to account for developmental reading problems in terms of a highly differentiated modular model derived from adult data. Interaction between levels creates serious dif culties in establishing direction of causation when multiple de cits are found in children with developmental disorders. Consider a hypothetical child who does poorly on auditory discrimination, vocabulary, and syntactic comprehension tests. When asked to distinguish between minimal pairs such as bear and pear, the child does much more poorly than other children of the same age. This is typically interpreted as evidence for an impairment of phonological discrimination. However, it may be that the child is less familiar with the vocabulary items, so that many of the test items are effectively treated as nonsense words. Nonsense words are usually more dif cult to discriminate than are real words. Thus the child’s weak vocabulary could exert a top-down in uence on performance on the discrimination task. And weak vocabulary could itself be a consequence of lack of syntactic knowledge, which would prevent the child from learning new word meanings by syntactic bootstrapping. In sum, it is not valid to assume that we can identify primary causes by locating the earliest point in the language comprehension chain at which a de cit can be seen. 3. Static Rather Than Developmental Models

In a developmental disorder, even if we do nd dissociations between de cits, these can be misleading, because the pattern of impairment may change over time. A goodexample is provided by a study by Bernstein and Stark (1985), who followed up a group of languageimpaired children who were originally tested by Tallal, Stark, Kallman, and Mellits (1981). Tallal and colleagues had demonstrated signi cant de cits in discriminating pairs of tones with brief interstimulus intervals. At follow-up several year later, many of these children were still impaired on a test of sentence comprehension. However, when Tallal’s test was readministered, the original de cit in discriminating tone pairs at short interstimulus intervals could no longer be seen; children with SLI did as well as control children. Thus at follow-up, we have a dissociation between intact auditory discrimination and impaired comprehension of sentences. Following conventional ``dissociation’’ logic, we might conclude that the language de cit in these children was not caused by any auditory dif culty. Yet we know from the earlier studies that several years previously the same children had marked impairments of auditory processing. This raises the possibility that a slow-maturing auditory perceptual system might leave a lasting legacy of language impairment, even after ceiling levels of auditory discrimination are reached. Other examples come from studies of individuals who have compensated for developmental disorders—that is, who have a clear history of reading or language impairments but who do not meet psychometric criteria for developmental dyslexia or SLI when seen in adolescence or adulthood. These individuals can often be revealed as having persisting dif culties with tests that require them to process unfamiliar materials (e.g. nonword reading or nonword repetition) (Bishop, North, & Donlan, 1996; Gross-Glenn, Jallad,

908

BISHOP

Novoa, Helgren-Lempesis, & Lubs, 1990; Stothard, Snowling, Bishop, Chipchase, & Kaplan, in press). Had we studied these individuals only at one late point in time, we might have concluded that the ability to process nonwords was irrelevant to language and literacy development. However, the past history tells a different story and suggests that the success in everyday language and reading tasks was achieved by adopting atypical compensatory strategies. For instance, a child who has dif culty in segmenting phonemes may learn to read laboriously by memorizing entire orthographic patterns. This strategy may be inef cient but ultimately successful. In adulthood, we see someone who has major problems in nonword reading but adequate real word reading: it would be mistaken, however, to use this observation to defend a model of normal reading that assumed no relationship between knowledge of grapheme_ phoneme correspondences and word recognition. The more longitudinal studies are carried out in the eld of SLI, the more evidence mounts for a changing pro le of language impairments with age, highlighting the dangers of basing a model of underlying processes solely on cross-sectional data. Furthermore, we have to take into consideration the fact that, in the course of development, the nature of representations may change. For example, in a review of research, Walley (1993) concluded that early phonological development involves a progression from larger to more ne-grained units of analysis. Initially, the child may operate with whole words or even short phrases, simply encoding these in terms of certain salient features, such as number of syllables, stress, and presence of phonetic features such as nasality or sibilance somewhere in the input (without any speci cation of exactly where) (Waterson, 1971, 1981). Such a child might not be able to distinguish between different syllables that contain the same features in different combinations but would nevertheless be able to store templates of words with incomplete representations of phonetic information. As Walley noted, `` Although seldom considered, the fact that children produce a certain contrast does not necessarily imply the presence of that contrast in their perceptual representations for words, which might instead be stored and retrieved as unsegmented wholes’’ (p. 317). By the age of 3 or 4 years, most children appear to be aware of the subsyllabic units of onset and rime, but only later, probably as a consequence of exposure to print, do they recognize the smaller phonemic elements. It has been suggested that the main impetus for moving to an analysis at the subsyllabic level is growth in vocabulary, which creates a need to impose a clear organization on lexical storage if words are to be recognized and retrieved rapidly (Jusczyk, 1986). Note that this example not only demonstrates the changing nature of representations, but also provides another instance of interaction between different levels of representation: growth in vocabulary is seen as in uencing the way in which phonological information is encoded. 4. Assumption that Modality of De® cit and Nature of Errors Relate to Primary Underlying Cause

In an adult with an acquired disorder, we can gain a good picture of the stage of processing that is affected by comparing tasks that involve similar representations in different input modalities, or by looking at the nature of errors. On the surface, it seems very reasonable to assume that similar logic applies to developmental disorders. However,


909

the case is less clear-cut than it may seem. For instance, there has been much argument concerning the interpretation of impairments in grammatical morphology in children with SLI. On the one hand, it has been noted that the dif culty experienced by children with SLI in learning the grammatical morphology in a given language is strongly related to the perceptual salience of in ectional endings, so that, in English, endings such as past tense ``-ed’’ are often omitted (see Bishop, 1997, for a review). However, against this view, others have noted that within a language (and indeed within the same child), the same word ending, such as ``-s’’, may be produced and perceived differentially depending on whether it functions as a plural, a possessive, or the nal phoneme in a word such as ``horse’’. Also, the same grammatical errors of both production and comprehension may be evident in written as well as spoken language (e.g., Bishop, 1982; Cromer, 1978). Authors such as Gopnik and Crago (1991) have regarded such ndings as evidence against auditory de cit accounts of SLI. The assumption is that, if the problem were simply that / s/ were dif cult to perceive, processing of all forms of / s/ should be impaired, and if auditory discrimination is the basis of the disorder, then much better performance should be observed with written presentation. However, this type of assumption is unsafe. Low-level general impairments of perceptual or cognitive systems can lead to unexpected patterns of de cit, both because of compensatory mechanisms on the one hand, and because of interactions between different components of a developing system on the other. If we take the extreme case of children with profound congenital hearing loss, on tests of grammatical comprehension of English, they perform very differently on items that simply require understanding of content words and on those that involve appreciating the signi cance of contrasts signalled by morphological endings, function words, and word order, regardless of whether or not these are perceptually salient (Bishop, 1982, 1983). Furthermore, presentation of sentences in a written form does very little to overcome these problems. Indeed, it was noted that there were some deaf children who could read aloud every word in a sentence such as the boy is not running or the man is pushed by the elephant but still selected the picture of a boy running and the man pushing the elephant. When confronted with complex syntactic constructions, deaf children did not just guess at random but, rather, adopted systematic strategies that sometimes led to below-chance performance. Even those children with facility in a native sign language had major dif culties in decoding the syntax of English—presumably because the surface forms used in oral languages to mark grammatical functions (i.e. word order and bound morphological af xes) are not well suited to processing in the visual modality, where simultaneous rather than sequential grammatical processes are the rule. Thus hearing loss did not simply lead to a slowing of language acquisition, nor did it lead to a predictable pattern of grammatical impairment affecting only non-salient morphemes. In fact, the distinctive response patterns seen in hearing-impaired children were remarkably similar to those seen in hearing children who had been diagnosed as having receptive language disorders. The crucial points emphasized by these studies of deaf children are that a general nonlinguistic impairment in auditory perception early in development can (a) have a disproportionately severe impact on the ability to produce and comprehend certain grammatical components of oral language (with different effects on grammatical morphology and vocabulary) and (b) affect written as well as spoken forms of an oral language. Note that this pattern of ndings is totally different from

910

BISHOP

what one will see if a hearing loss was acquired in adulthood, after pro ciency in oral language had developed. Hearing loss does not just disrupt the input to speci c parts of the language system; if acquired early in life, it alters the entire course of oral language development. It must be stressed that I am not suggesting that SLI is caused by hearing loss. Rather, the point is that it is dangerous to imagine that we can predict on a priori grounds what pattern of language impairment would result if the child had a nonspeci c cognitive or perceptual de cit that affected language learning—for example, a disorder of auditory perception such as has been proposed by Tallal (1976) or a limitation of phonological short-term memory as mooted by Gathercole and Baddeley (1990). The impact of a de cit at an early stage of processing can be felt across the whole language system. A further example comes from the domain of reading disability. Castles and Coltheart (1993) argued that error patterns could be used to identify which route of a dual-route reading system was impaired in developmental cases, just as had been done for acquired dyslexia. They produced evidence for developmental forms of two contrasting subtypes: surface and phonological dyslexia, which together gave a double dissociation. Surface dyslexia is de ned as poor ability to read exception words relative to nonwords, whereas phonological dyslexics show the opposite pattern. However, Snowling and Nation (1997) found in a longitudinal study that not only did many poor readers show characteristics of both surface and phonological dyslexia, but that they moved from one subtype to another over a two-year period. Stanovich, Siegel, and Gottardo (1997) threw some light on such paradoxical ndings in a reanalysis of data from Castles and Coltheart, boosted by further data from a younger sample of poor readers. They found that the surface dyslexic error pattern was typical for normally developing children at an early stage of reading ability— that is, it could be regarded as a developmental lag rather than indicating impaired development of one route to reading. Furthermore, Stanovich and colleagues argued that both subtypes of reading disability might lie on a continuum, with the precise manifestation depending on the severity of the underlying phonological impairment and the amount of exposure to printed words that the child received. Thus the error pattern that one observes may re ect the child’s history of reading exposure as well as underlying cognitive strengths and weaknesses. 5. Modules as Hard-wired Innate Systems

Fodor’s de nition of a module (Table 1) includes a number of different features that are logically separable. He proposes that if a cognitive process has the characteristics of domain-speci city and information encapsulation, then it is likely to be an evolutionary adaptation with an innate biological basis. Those postulating modular explanations of SLI have emphasised the mounting evidence, from both pedigree and twin studies, that SLI has a genetic basis, and in some cases may even be caused by a single defective gene (North & Donlan, 1995; Gopnik & Crago, 1991; Hurst, Baraitser, Auger, Graham, & Norrell, 1990; Lewis & Thompson, 1992; Tomblin & Buckwalter, 1994). On the basis of such evidence, Gopnik and Crago (1991) made the tentative proposal that a single gene might control mechanisms responsible for learning morphological paradigms. However, accounts of this work in secondary sources have made more extravagant claims, not only


911

over-exaggerating the selectivity of the morphological de cit (Gazzaniga, 1992, cf. Vargha-Khadem, Watkins, Alcock, Fletcher, & Passingham, 1995), but also treating the genetic etiology of SLI as evidence for an innate grammatical module. Thus, Jackendoff (1993) argued that: these results strongly suggest that the impairment is genetic, and that it speci cally affects the ability to construct a mental grammar, leaving other cognitive abilities intact. In order for this to be possible, there must be at least one gene that is responsible for a special-purpose mental endowment for language acquisition. The part of Universal Grammar having to do with acquiring in ectional endings must not be a general-purpose learning strategy (p. 116).

The danger of this logic may be illustrated with a simple analogy between the ability to walk and the ability to talk. Like language, walking is a ``species universal’’—that is, common to all normal humans but not seen in other primates—which develops without overt instruction in a wide range of environmental circumstances. The ability to walk upright across an uneven terrain involves astonishingly complex co-ordination of motor and proprioceptive systems. No other primate shares this skill, yet all normal humans master it without speci c instruction within the rst few years of life, regardless of whether their local environment is a desert plain, a rainforest, a snowy mountainside, or a Western city. Attempts to build bipedal robots have met with some success, but the end results still fall far short of a typical 3-year-old’s competence. Just as with language, though, some unfortunate individuals have a speci c single-gene disorder—muscular dystrophy—which selectively interferes with this ability. However, nobody supposes that because a single genetic mutation can cause muscular dystrophy, there is a ``gene for walking’’. It is apparent that walking depends on the integrity of a wide range of underlying systems, involving muscles, nerves, and central control processes that regulate balance, proprioception, and motor planning. Muscular dystrophy has a speci c effect on just one of these systems—the muscles— but this is suf cient to make walking dif cult or impossible. Those favouring processing accounts of SLI would argue that just the same can be said for language. The path from gene to behaviour is an immensely complex one, and there are many possible explanations for SLI. Genes associated with SLI may operate by disturbing the development of those brain areas concerned with analysing or retaining transient auditory stimuli, for instance. Fodor argued that functions that met the rst few criteria for modularity outlined in Table 1 were likely to be ``hard-wired’’ and innate, so that they could function rapidly and reliably without requiring a prolonged learning period. However, models of developmental processes suggest a very different story, whereby ``modularity’’, in the sense of a domain-speci c and informationally encapsulated system, emerges with experience (see Elman et al., 1996; Tucker & Hirsh-Pasek, 1993). The distinction between `àutomatic’’ and ``controlled’’ processes, which maps closely onto Fodor’s modular/ central processes distinction, seems more a function of practice and experience than of task domain. Indeed, some highly automatic processses involve arti cial tasks that could not plausibly be regarded as innate, such as piano playing or bicycle riding. Nor does localization of a function in a particular brain region imply innateness. There is plentiful evidence from acquired dyslexia that the brain regions involved in reading are

912

BISHOP

highly localized and specialized, but reading is a relatively late arrival on the stage of world history, and the majority of the world’s population does not read and write. Cerebral specialization depends as much on speci c experiences as on pre-wiring. The fact, then, that in adults certain aspects of language processing have the characteristics of modularity—domain speci city, information encapsulation, and localization—tells us nothing about whether the brain is innately set up for this type of processing, and the nding of a genetic basis of SLI is compatible with a de cit in an innate grammatical module, but it is not evidence for such an account.

II:

Problems for the Subject in General 6. Emphasis on Representational Rather than Processing De® cits

Suppose a person is poor at producing grammatical sentences and often omits in ectional endings. One way of accounting for this would be to propose that some level of grammatical representation was defective—namely, that there is an impairment of language competence. An alternative account would be to argue that there is a processing de cit, in which linguistic performance is constrained by a slow rate of access to representations or rapid decay in memory. According to this model, the underlying representations may be intact, but there would be problems in deploying grammatical knowledge under real time constraints. This kind of model would be more akin to the accounts put forward to explain the speech errors made by normal individuals: we all make ``slips of the tongue’’, but we do not assume that we have lost grammatical knowledge. In the eld of SLI there is considerable controversy between those who assume that syntactic de cits re ect a lack of linguistic knowledge (e.g. Clahsen, 1989; Gopnik & Crago, 1991; van der Lely, 1994), and those who propose explanations in terms of limited processing capacity (e.g., Chapman, 1992; Gathercole & Baddeley, 1990; Johnston, 1994; Tallal, Miller, & Fitch, 1993). Models with the box-and-arrows architecture of Figure 1 are useful for describing representational de cits, but they are less well suited for characterizing processing de cits. Reliance on such models has probably played a part in the relative neglect of processing accounts of language disorders in cognitive neuropsychology, despite evidence that the comprehension of aphasic patients could be in uenced by factors unrelated to linguistic content, such as rate of presentation (e.g. Albert & Bear, 1974). A notable exception is the study of semantic de cits in acquired aphasia and dyslexia, where there is lively debate about how best to distinguish between a patient with degraded semantic representations and one who has intact representations but dif culty in accessing these (e.g. Shallice, 1988; Warrington & Shallice, 1979). More recently, some cognitive neuropsychologists have attempted to incorporate processing factors in models of syntactic impairment (e.g. Haarman & Kolk, 1991), but in general the tendency is still to look for explanations in terms of de cient linguistic representations. How are we to decide between representational and processing accounts? There are both similarities and differences in the approaches adopted by those dealing with developmental versus acquired disorders. In discussing semantic access disorders in adults,


913

Shallice (1988) notes that inconsistent performance is an indication that underlying representations are intact but not adequately accessed. In a similar vein Bishop (1994b) noted that if a child shows partial mastery of a grammatical structure—that is, performance that is above chance but still defective, this poses problems for a representational account of grammatical de cits, because, according to such theories, a child either does or does not know a grammatical rule. In both acquired and developmental elds, there is growing recognition that one often sees a level of performance that is intermediate between unimpaired and random, and hence dif cult to explain in terms of defective representations. However, we must always remember that test performance is only an indirect indicator of underlying cognitive operations and be aware of the possibility that good performance, as well as errors, may misrepresent the individual’s capabilities. For instance, Goad and Rebellati (1994) noted that some individuals with SLI did show some ability to pluralize nonsense words, a skill that is typically interpeted as prima facie evidence that the person has knowledge of productive morphological rules. However, they suggested that these individuals were explicitly applying the taught rule: `àdd -s to form a plural’’. Not only did this lead to slow and effortful production, but detailed phonetic analysis showed that the apparently correct plural ``-s’’ forms produced by languageimpaired individuals lacked voicing assimilation: for instance, when asked for the plural of ``wug’’, the response was / WL gS/ rather than / WL gZ/ . The point stressed by these researchers is that correct performance on a language test need not necessarily re ect true competence in underlying ability. Grammatical morphemes might be used without appreciation of their syntactic signi cance, being applied consciously after explicit instruction in their use. A second crucial feature of processing theories is that they predict that the ability to do linguistic tasks will vary depending on task demands. In the study of semantic de cits in aphasia, it has been demonstrated that some patients do much better in tests of comprehension or naming if a long interval (e.g. 30 seconds) elapses between each response and the next trial, than if a standard presentation rate (e.g., 1 to 2 seconds between items) is used (Forde & Humphreys, 1995; McNeil, Cipolotti, & Warrington, 1994; Warrington & McCarthy, 1983). Rather less has been done on the effect of presentation rate on grammatical processing, although comprehension of complex sentences has been shown to be impaired in normal adults at rapid presentation rates (Miyake, Carpenter, & Just, 1994). In a study of children with SLI, McCroskey and Thompson (1973) showed that for younger children, performance on a sentence comprehension task improved signi cantly when the sentences were presented at a rate of 2.9 syllables per second, compared with 5.0 syllables per second, indicating that the children had the necessary linguistic knowledge to do the task but were poor at deploying it under real time constraints. One factor that has been explored in the developmental context is the notion of tradeoffs between different levels of processing. Because cognitive neuropsychology adopts a modular framework, the possibility of interactions between different levels of processing is seldom considered, but from a developmental perspective one would expect them. Trade-offs make sense if we assume that there is a limited-capacity system that handles several linguistic operations simultaneously, so that complexity at one level can lead to errors at another. For instance, if we restrict ourselves to vocabulary that the child knows but include some words that are relatively infrequent and others that are common, can we

914

BISHOP

in uence the dif culty of a grammatical comprehension test by in uencing the ease of lexical access? Does phonological complexity or word length affect the ability to understand a complex sentence? If we nd interactions between language levels, this would be supportive of a processing explanation of comprehension dif culties. To my knowledge, experimental studies of this kind have not been conducted on comprehension in children with SLI, but evidence for trade-offs has been found in analyses of children’s language output (Crystal, 1987; Masterson & Kamhi, 1992; Panagos & Prelock, 1982). Notions of limited processing capacity make testable predictions that ability to produce and comprehend language will depend not just on the nature and complexity of the representations that are accessed, but also on the overall processing demands placed on the individual. In contrast to the modular approach derived from cognitive neuropsychology, they steer us away from looking at different components of language in isolation and draw attention to the trade-offs that may exist between different levels of processing and to the effects of variables such as speech rate on linguistic processing. Processing accounts predict that we may be able to manipulate level of performance with particular sentence structures by changing the processing demands of a task. If so, this would not only further our theoretical understanding of SLI, but would also have implications for intervention. 7. Problems in Interpreting Differential De® cit

The textbook case of a patient who is entirely normal on tasks measuring one set of cognitive operations (X) and completely unable to do another set (Y) is very rarely observed, even if one restricts consideration to acquired cases with focal lesions. More often, arguments in cognitive neuropsychology are based on evidence of differential de cit—that is, the patient might be impaired in both X and Y, but the impairment in X is more severe. However, as Wilding (1989) pointed out, researchers seldom make explicit their criteria for deciding that a difference between X and Y should be regarded as a signi cant dissociation. When one attempts to specify such criteria, one comes up against methodological dif culties related to psychometric characteristics of tests, especially item dif culty and item reliability. Most researchers are aware of interpretative problems created by oor and ceiling effects. For instance, in Figure 3, the control and SLI groups differ signi cantly on Test X but not on Test Y, but we would be wary of assuming that there was differential de cit in X because performance on Y is near ceiling for both groups, and hence by choosing too easy a test we may have masked a real difference between groups. What is much less well appreciated is that even when group performance is off oor or ceiling, differences in the relative dif culty of items in two tests can lead to spurious ``dissociations’’. Chapman and Chapman (1973) recommended that researchers attempt to match the dif culty level of tasks that are to be compared and use items that give good discrimination within a control group wherever possible. They also illustrated the extent to which test reliability can in uence the size of differences in scores between two groups. Figure 4 shows real data from a multiple-choice analogies test obtained by these authors from 49 people with schizophrenia and 206 normal controls. From within the same test, they selected items to form a high-reliability (X) and a lowreliability (Y) version. Only the more reliable version showed a signi cant de cit in the


915

FIG. 3. Differential de cit complicated by a ceiling effect ( ctitious data).

group with schizophrenia. The point emphasized by Chapman and Chapman is that the two tests were selected to be equivalent both in terms of item content and in terms of overall dif culty in the control group. It is clearly the difference in test reliability that is reponsible for the differential de cit of the schizophrenic group. Yet, if X and Y had different types of content, we might easily have been misled into assuming that the people with schizophrenia had a selective de cit in carrying out the cognitive operations involved in Test X. In general, the demonstration of differential de cit is less of a problem if our control task (Y) is a well-standardized psychometric instrument, because it is then reasonable to suppose that Y is at least as discriminating and reliable as an experimental test (X), and so if we nd that people who are matched on Yperform differently on X, it is unlikely that this is because Y is less sensitive. However, where both X and Y are assessed by experimental tasks, Chapman and Chapman recommended that researchers match the

FIG. 4. Effect of test reliability on sensitivity to de cit. Data from Chapman and Chapman (1973). Reliability is assessed in terms of coef cient alpha, a measure of how well individual items correlate with the test total (i.e. internal consistency).

916

BISHOP

tasks on discriminating power and reliability, in order to have con dence in interpreting dissociations. In classic cognitive neuropsychology, these kinds of dif culty are implicitly recognized by those who stress the particular importance of double dissociations for informing theory (Milner & Teuber, 1968). In short, if we have two patients who show the opposite patterns of de cit, with one selectively impaired on X and the other selectively impaired on Y, then it would seem that we are in a stronger position to conclude that X and Yare independent faculties and to reject any simple explanation in terms of ``general’’ task factors, such as test reliability or dif culty. Even here, however, some awareness of psychometric principles is needed. Bates, Appelbaum, and Allard (1991) noted that some dissociations between measures can be expected by chance alone (this being particularly likely when test reliability is low and when reliance is placed on data from a single case).

BEYOND COGNITIVE NEUROPSYCHOLOGY Studying Interaction and Change in Language Impairments Given the dif culties encountered when trying to apply the traditional methods of cognitive neuropsychology to a developmental disorder, what are the alternative approaches one can adopt? The traditional kind of study, where children with SLI are compared with a control group on one or more language tests, is usually a useful rst step in trying to elucidate the nature of SLI, but it is unlikely to resolve questions about primacy of speci c cognitive de cits. Striking dissociations are rare in children, and even when they occur, their interpretation is often ambiguous because of changing patterns of impairment with age. Associations between de cits are even more dif cult to interpret, because they could re ect bottom-up in uences on high-level processes, top-down in uences on low-level processes, or it could be that the de cits are causally unconnected but tend to co-occur because they are mediated by adjacent brain areas. This does not mean that we should abandon attempts at furthering our understanding of psycholinguistic processes in SLI, only that we need to recognize that converging evidence from different methodologies will be needed. To conclude this paper, a brief overview of different methods is offered for consideration. Language Age-Matched Controls

One of the commonest strategies used in the study of SLI is to compare a languageimpaired group with a younger control group matched on some index of ``language age’’. The rationale for this approach is that it should give us some indication of whether a speci c impairment on a test, X, is disproportionate to the child’s language level. If not, it could be argued that poor performance on X is simply a consequence of low language skills rather than being of primary importance in causing the language dif culties. However, if children with SLI do even more poorly than language-matched controls, this indicates that we can’t just dismiss the de cit as secondary.


917

There are several dif culties inherent in this approach. First, it is an exceedingly conservative method. A lack of difference between groups does not tell us much except that the impairment could be a secondary one. As the children with SLI are older than the control group, they are likely to be more advanced in general cognitive development and to have better attention and a better ability to develop strategies for doing an experimental task, and these could counteract any effect of underlying impairment. Second, `àgeequivalent’’ scores can vary substantially from one test to another, and it can be dif cult to know which language test children should be matched on (see Bishop, 1997). Third, the psychometric issues involved in looking for differential de cit (see earlier) are critical in this kind of design but are often ignored. Comparison with Other Groups with Known Level of Impairment

When considering the impact of relatively peripheral impairments on language development—that is, those affecting sensory or motor processes, useful insights can be obtained by studying children suffering from conditions known to affect these. Bishop’s (1982, 1983) studies of comprehension of English grammar in children with hearing impairment provide one example. Contrary to expectation, these studies revealed a distinctive pattern of grammatical dif culties, cutting across input modality, in such children and thus offered indirect supporting evidence to those who would argue that an early auditory impairment can distort acquisition of oral syntax. Another example comes from studies of children who are dysarthric or anarthric and therefore nd oral speech dif cult or impossible because of impairments of motor control. These children allow one to address questions such as how far articulatory coding plays a role in memory or phonological segmentation (e.g. Bishop & Robson, 1989a, 1989b) and whether experience of producing language is critical for the development of understanding (Bishop, Byers-Brown, & Robson, 1990). In general, these studies have revealed surprisingly intact abilities in domains of phonological and syntactic processing in children who cannot produce clear speech, and they therefore make us less ready to assume that comprehension dif culties in children with SLI are due to their expressive limitations. Experimental Studies on ``Trade-offs’’ and Capacity Limitations

A growing interest in processing accounts of language development and disorders encourages us to do more studies using a within-subjects design, where the goal is to see how various experimental manipulations affect the child’s ability to understand, as opposed to the more traditional between-subjects studies, where two or more groups are compared. Methods include investigating the effects of varying such factors as rate of presentation, background noise, prosody, and so on, on children’s understanding, as well as, within the same task, varying dif culty at different levels of processing, such as phonology, semantics, and syntax. A nice feature of such studies is the potential they have for providing results that are useful in intervention, by specifying which parameters make it easier or more dif cult for children to process language.

918

BISHOP

Longitudinal Studies

The potential importance of longitudinal studies for addressing causal questions has already been emphasized. Such methods have not been widely used in the study of SLI, and application of sophisticated statistical modelling (see, for instance, Magnusson, Bergman, Rudinger, & To¨restad, 1989, for examples) has barely been attempted. Nevertheless, even with a fairly simple methodological approach, longitudinal studies can be invaluable in helping clarify the direction of causal relationships in normal development (e.g. Gathercole, Willis, Emslie, & Baddeley, 1992), for highlighting the role that early impairments may play in causing later de cits (e.g. Bernstein & Stark, 1985) and for throwing light on questions of classi cation (see, e.g., Bishop, 1994c; Bishop&Edmundson, 1987). Modelling the Learning Process and Simulating Effects of Impairment at a Given Stage

Newdevelopments in dynamic systems theory (Thelen & Smith, 1993) and connectionist modelling (Elman et al. 1996) are making it possible to develop simulations of developmental change, in contrast to previous computational approaches which were based on static information-processing systems. These have the potential to let us see how the nature of representations changes as an organism interacts with its environment and how uctuations either in the biological substrate or in the input it receives may affect developing language processes. Intervention Studies

The ultimate test of a hypothesis is through experimental manipulation. If one believes that one has identi ed the primary process that is implicated in SLI, then by ameliorating that de cit, one should be able to show bene cial effects on other aspects of language development. For instance, Tallal et al. (1996) exposed language-impaired children to acoustically modifed speech in which the duration of the speech signal was prolonged and the transitional elements were ampli ed. Control children received a similar package of interventions, but using unmodi ed speech and computer games based on discriminations that did not involve rapid processing. The outcomes reported by Tallal and colleagues were dramatic, with children who received the ``temporal processing’’ intervention showing signi cantly greater improvement than control children on language measures. This work, which has excited considerable controversy, is still at an early phase, and further evaluation studies are under way. It remains unclear which components of the intervention package are critical. If the early promise of the method is ful lled, with auditory training producing improved language abilities in everyday contexts, this would be the strongest support yet for the view that linguistic dif culties in these children are secondary to more fundamental auditory limitations. Another example comes from a study brie y described by Gillam and van Kleek (1996). They found that training language-impaired children in phonological awareness


919

led to improvements in their nonword repetition without in uencing real word repetition. This supports the view that poor nonword repetition in children with SLI may have more to do with how phonological information is initially encoded in memory than with limited storage capacity or rapid decay of stored information (see Bishop, 1997). Although applications to intervention are frequently cited by researchers as the justi cation for doing experimental studies, all too often the link with clinical practice is never made. It is time for researchers to recognize that intervention studies are not just an optional, applied adjunct to experimental work, but that they provide the best method available for evaluating hypotheses and unconfounding correlated factors.

Conclusion It must be stressed that I am not saying that cognitive neuropsychology has nothing to offer developmental psychology. Case studies of adults with acquired lesions can provide important insights into the stages of processing that are involved in a particular cognitive process, and they can lead to the development of ingenious tasks for pinpointing the underlying cause of impaired language performance. However, it is dangerous to assume that a model of cognitive processing that is derived from the study of adults can be applied without modi cation to children: there is ample evidence that the nature of underlying representations may evolve in the course of development, and there may be far more interaction between levels of processing in children than adults. When interpreting dissociations, one needs to be aware that the pro le of impairment can change with age and that an impairment affecting one stage of processing may have bottom-up or top-down in uences on the development of other stages. If the same child who is classi ed as a ``surface dyslexic’’ at one point in time turns into a ``phonological dyslexic’’ later in development, then this questions the rationale of treating these as complementary disorders involving impairment to different pathways. Furthermore, test sensitivity and reliability will in uence how easy it is to show differential de cit: if we nd a child who is impaired in Test X but not in Test Y, we tend to assume that this is because X is an index of some underlying cognitive process (A) that is compromised, whereas the function, B, measured by Test Y, is still intact. However, as argued above, it could be that X is just a more sensitive, reliable and/ or valid index of A than Test Y as an index of B. The fact that different cognitive functions may show different growth trends in the normal population complicates the picture yet further: Test X may be an excellent index of A in 6-year-olds but be insensitive for 12-year-olds. Finally, it is important to distinguish whether our primary goal is to use disorders to understand normal development or to develop a model of the disorder itself. If our aim is to increase our understanding of SLI, we need to focus not so much on demonstrating dissociations as on interpreting complex patterns of impairment. In this context, the single case approach is singularly inappropriate, because we cannot tell whether the cooccurrence of impairments is a chance occurrence or a systematic pattern. We need to study groups of individuals in order to look at correlations between impairments in different processes.

920

BISHOP

REFERENCES Albert, M.L., & Bear, D. (1974). Time to understand: A case study of word deafness with reference to the role of time in auditory comprehension. Brain, 97, 373_ 384. Baddeley, A., Papagno, C., & Vallar, G. (1988). When long-term learning depends on short-term storage. Journal of Memory and Language, 27, 586_ 595. Bates, E., Appelbaum, M., & Allard, L. (1991). Statistical constraints on the use of single cases in neuropsychological research. Brain and Language, 40, 295_ 329. Berndt, R.S., Basili, A., & Caramazza, A. (1987). Dissociation of functions in a case of transcortical sensory aphasia. Cognitive Neuropsychology, 4, 79_ 107. Bernstein, L.E., & Stark, R.E. (1985). Speech perception development in language-impaired children: A 4-year follow-up study. Journal of Speech and Hearing Research, 50, 21_ 30. Bishop, D.V.M. (1982). Comprehension of spoken, written and signed sentences in childhood language disorders. Journal of Child Psychology and Psychiatry, 23, 1_ 20. Bishop, D.V.M. (1983). Comprehension of English syntax by profoundly deaf children. Journal of Child Psychology and Psychiatry, 24, 415_ 434. Bishop, D.V.M. (1992). The underlying nature of speci c language impairment. Journal of Child Psychology and Psychiatry, 33, 1_ 64. Bishop, D.V.M. (1994a). Developmental disorders of speech and language. In M. Rutter, L. Hersov, & E. Taylor (Eds.), Child and adolescent psychiatry (pp. 546_ 568). Oxford: Blackwell Scienti c. Bishop, D.V.M. (1994b). Grammatical errors in speci c language impairment: Competence or performance limitation? Applied Psycholinguistics, 15, 507_ 549. Bishop, D.V.M. (1994c). Is speci c language impairment a valid diagnostic category? Genetic and psycholinguistic evidence. Philosophical Transactions of the Royal Society, Series B, 346, 105_ 111. Bishop, D.V.M. (1997). Uncommon understanding: Development and disorders of language comprehension in children. Hove: Psychology Press. Bishop, D.V.M., Byers-Brown, B., & Robson, J. (1990). The relationship between phoneme discrimination, speech production and language comprehension in cerebral-palsied individuals. Journal of Speech and Hearing Research, 33, 210_ 219. Bishop, D.V.M., & Edmundson, A. (1987). Language-impaired four-year-olds: Distinguishing transient from persistent impairment. Journal of Speech and Hearing Disorders, 52, 156_ 173. Bishop, D.V.M., North, T., & Donlan, C. (1995). Genetic basis of speci c language impairment: Evidence from a twin study. Developmental Medicine and Child Neurology, 37, 56_ 71. Bishop, D.V.M., North, T., & Donlan, C. (1996). Nonword repetition as a behavioural marker for inherited language impairment: Evidence from a twin study. Journal of Child Psychology and Psychiatry, 37, 391_ 403. Bishop, D.V.M., & Robson, J. (1989a). Unimpaired short-term memory and rhyme judgement in congenitally speechless individuals: Implications for the notion of `àrticulatory coding’’ . Quarterly Journal of Experimental Psychology, 41A, 123_ 140. Bishop, D.V.M., & Robson, J. (1989b). Accurate nonword spelling despite congenital inability to speak: Phoneme_ grapheme conversion does not require subvocal articulation. British Journal of Psychology, 80, 1_ 13. Caramazza, A. (1986). On drawing inferences about the structure of normal cognitive systems from the analysis of patterns of impaired performance: The case for single-patient studies. Brain and Cognition, 5, 41_ 66. Castles, A., & Coltheart, M. (1993). Varieties of developmental dyslexia. Cognition, 47, 149_ 180. Chapman, L.J., & Chapman, J.P. (1973). Problems in the measurement of cognitive de cit. Psychological Bulletin, 79, 380_ 385. Chapman, R.S. (1992). Processes in language acquisition and disorders. St Louis, MO: Mosby Year Book. Clahsen, H. (1989). The grammatical characterization of developmental dysphasia. Linguistics, 27, 897_ 920. Coltheart, M. (1987). Functional architecture of the language processing system. In M. Coltheart, G. Sartori, & R. Job (Eds.), The cognitive neuropsychology of language (pp. 1_ 25). London: Lawrence Erlbaum Associates Ltd.


921

Crago, M.B., & Gopnik, M. (1994). From families to phenotypes: Theoretical and clinical implications of research into the genetic basis of speci c language impairment. In R. Watkins & M. Rice (Eds.), Speci c language impairments in children (pp. 35_ 51). Baltimore, MD: Paul H. Brookes. Cromer, R.F. (1978). The basis of childhood dysphasia: A linguistic approach. In M. A. Wyke (Ed.), Development dysphasia (pp. 85_ 134). London: Academic Press. Crystal, D. (1987). Towards a ``bucket’’ theory of language disability taking account of interaction between linguistic levels. Clinical Linguistics and Phonetics, 1, 7_ 22. Elman, J., Bates, E., Johnson, M.H., Karmiloff-Smith, A., Parisi, D., & Plunkett, K. (1996). Rethinking innateness: A connectionist perspective on development. Cambridge, MA: MIT Press. Fodor, J.A. (1983). The modularity of mind. Cambridge, MA: Bradford Books. Forde, E., & Humphreys, G.W. (1995). Refractory semantics. Memory, 3, 265_ 307. Gardner, H., & Zurif, E. (1975). BEE but not BE: Oral reading of single words in aphasia and alexia. Neuropsychologia, 13, 181_ 190. Gathercole, S.E., & Baddeley, A.D. (1990). Phonological memory de cits in language disordered children: Is there a causal connection? Journal of Memory and Language, 29, 336_ 360. Gathercole, S.E., Willis, C., Emslie, H., & Baddeley, A.D. (1991). The in uences of syllables and wordlikeness on children’s repetition of nonwords. Applied Psycholinguistics, 12, 349_ 367. Gathercole, S.E., Willis, C., Emslie, H., & Baddeley, A.D. (1992). Phonological memory and vocabulary development during the early school years: A longitudinal study. Developmental Psychology, 28, 887_ 898. Gazzaniga, M.S. (1992). Nature’s mind. London: Penguin Books. Gillam, R.B., & Van Kleek, A. (1996). Phonological awareness training and short-term working memory: Clinical implications. Topics in Language Disorders, 17, 72_ 81. Gleitman, L.R. (1994). Words words words. Philosophical Transactions of the Royal Society, Series B, 346, 71_ 77. Goad, H., & Rebellati, C. (1994). Pluralization in familiar language impairment. In J. Matthews (Ed.), Linguistic aspects of familial language impairment: Special issue of the McGill working papers in linguistics, Vol. 10 (pp. 24_ 40). Goldberg, E. (1995). Rise and fall of modular orthodoxy. Journal of Clinical and Experimental Neuropsychology, 17, 193_ 208. Gopnik, M., & Crago, M. (1991). Familial aggregation of a developmental language disorder. Cognition, 39, 1_ 50. Gross-Glenn, K., Jallad, B., Novoa, L., Helgren-Lempesis, V., & Lubs, H.A. (1990). Nonsense passage reading as a diagnostic aid in the study of adult familial dyslexia. Reading and Writing, 2, 161_ 173. Haarman, H.J., & Kolk, H.H.J. (1991). A computer model of the temporal course of agrammatic sentence understanding: The effects of variation in severity and sentence complexity. Cognitive Science, 15, 49_ 87. Hulme, C., & Snowling, M. (1992). De cits in output phonology: An explanation of reading failure? Cognitive Neuropsychology, 9, 47_ 72. Hurst, J.A., Baraitser, M., Auger, E., Graham, F., & Norell, S. (1990). An extended family with a dominantly inherited speech disorder. Developmental Medicine and Child Neurology, 32, 352_355. Jackendoff, R. (1993). Patterns in the mind. Hemel Hempstead: Harvester Wheatsheaf. Johnston, J. (1994). Cognitive abilities of children with language impairment. In R. Watkins & M. Rice (Eds.), Speci c language impairments in children (pp. 107_ 121). Baltimore: Paul H. Brookes. Jusczyk, P. (1986). Toward a model of the development of speech perception. In J.S. Perkell & D.H. Klatt (Eds.), Invariance and variability in speech processes (pp. 1_ 19). Hillsdale NJ: Lawrence Erlbaum Associates Ltd. Landau, B., & Gleitman, L. (1985). Language and experience: Evidence from the blind child. Cambridge, MA: Harvard University Press. Lewis, B.A., & Thompson, L.A. (1992). A study of developmental speech and language disorders in twins. Journal of Speech and Hearing Research, 35, 1086_ 1094. Magnusson, D., Bergman, L.R., Rudinger, G., & To¨restad, B. (1989). Problems and methods in longitudinal research: Stability and change. Cambridge: Cambridge University Press. Marslen-Wilson, W., & Tyler, L. (1987). Against modularity. In J.L. Gar eld (Ed.), Modularity in

922

BISHOP

knowledge representation and natural-language understanding (pp. 37_ 62). Cambridge, MA: MIT Press. Masterson, J.J., & Kamhi, A.G. (1992). Linguistic trade-offs in school-age children with and without language disorders. Journal of Speech and Hearing Research, 35, 1064_ 1075. McCroskey, R.L., & Thompson, N.W. (1973). Comprehension of rate controlled speech by children with speci c learning disabilities. Journal of Learning Disabilities, 6, 621_ 628. McNeil, J.E., Cipolotti, L., & Warrington, E.K. (1994). The accessibility of proper names. Neuropsychologia, 32, 193_ 208. Milner, B., & Teuber, H.-L. (1968). Alterations of perception and memory in man: Re ections on method. in L. Weiskrantz (Ed.), Analysis of behavioral change (pp. 268_ 375). New York: Harper and Row. Miyake, A., Carpenter, P.A., & Just, M.A. (1994). A capacity approach to syntactic comprehension disorders: Making normal subjects perform more like aphasic patients, Cognitive Neuropsychology, 11, 671_ 717. Panagos, J.M., & Prelock, P.A. (1982). Phonological constraints on the sentence productions of language-disordered children. Journal of Speech and Hearing Research, 25, 171_ 177. Saffran, E.M., Marin, O.S.M., & Yeni-Komshian, G.H. (1976). An analysis of speech perception in word deafness. Brain and Language, 3, 209_ 228. Schwartz, M., Saffran, E.M., & Marin, O.S.M. (1980). The word order problem in agrammatism. I. Comprehension. Brain and Language, 10, 249_ 262. Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press. Snowling, M., & Nation, K. (1997). Language, phonology and learning to read. In C. Hulme & M. Snowling (Eds.), Dyslexia: Biology, cognition and intervention. London: Whurr Publishers. Stanovich, K.E., Siegel, L.S., & Gottardo, A. (1997). Progress in the search for dyslexia sub-types. In C. Hulme & M. Snowling (Eds.), Dyslexia: Biology, cognition and intervention. London: Whurr Publishers. Stothard, S.E., Snowling, M.J., Bishop, D.V.M., Chipchase, B.B., & Kaplan, C.A. (in press). Languageimpaired preschoolers: A follow-up into adolescence. Journal of Speech, Language, and Hearing Research. Tallal, P. (1976). Rapid auditory processing in normal and disordered language development. Journal of Speech and Hearing Research, 19, 561_ 571. Tallal, P., & Katz, W. (1989). Neuropsychological and neuroanatomical studies of developmental language/ reading disorders: Recent advances. In C. von Euler, I. Lundberg, & G. Lennerstrand (Eds.), Brain and reading (pp. 183_ 196). New York: Stockton Press. Tallal, P., Miller, S.L., Bedi, G., Byma, G., Wang, X., Najarajan, S.S., Schreiner, C., Jenkins, W.M., & Merzenich, M.M. (1996). Language comprehension in language-learning impaired children improved with acoustically modi ed speech. Science, 271, 81_ 84. Tallal, P., Miller, S., & Fitch, R.H. (1993). Neurobiological basis of speech: A case for the pre-eminence of temporal processing. In P. Tallal, A.M. Galaburda, R.R. Llinas, & C. von Euler (Eds.), Temporal information processing in the nervous system (Annals of the New York Academy of Sciences, Vol. 682, pp. 27_ 47). New York: New York Academy of Sciences. Tallal, P., Stark, R., Kallman, C., & Mellits, D. (1981). A re-examination of some nonverbal perceptual abilities of language-impaired and normal children as a function of age and sensory modality. Journal of Speech and Hearing Research, 24, 351_ 357. Temple, C.M. (1992). Developmental dyscalculia. In S.J. Segalowitz & I. Rapin (Eds.), Handbook of neuropsychology, Vol. 7, Section 10: Child neuropsychology, Part 2 (pp. 211_ 222). Amsterdam: Elsevier Science. Temple, C.M. (1997). Developmental cognitive neuropsychology. Journal of Child Psychology and Psychiatry, 38, 27_ 52. Thelen, E., & Smith, L.B. (1993). A dynamic systems approach to the development of cognition and action. Cambridge, MA: MIT Press. Tomblin, J.B., & Buckwalter, P.R. (1994). Studies of genetics of speci c language impairment. In R. Watkins & M. Rice (Eds.), Speci c language impairments in children (pp. 17_ 34). Baltimore, MD: Paul H. Brookes. Tucker, M., & Hirsh-Pasek, K. (1993). Systems and language: Implications for acquisition. In L.B.


923

Smith & E. Thelen (Eds.), A dynamic systems approach to development: Applications (pp. 359-384). Cambridge, MA: MIT Press. Vallar, G., & Baddeley, A.D. (1984). Phonological short-term store, phonological processing and sentence comprehension: A neuropsychological case study. Cognitive Neuropsychology, 1, 121_ 141. Van der Lely, H.K.J. (1994). Canonical linking rules: Forward versus reverse linking in normally developing and speci cally language-impaired children. Cognition, 51, 29_ 72. Vargha-Khadem, F., Watkins, K., Alcock, K., Fletcher, P., & Passingham, R. (1995). Praxic and nonverbal cognitive de cits in a large family with a genetically transmitted speech and language disorder. Proceedings of the National Academy of Sciences, 92, 930_ 933. Walley, A.C. (1993). The role of vocabulary development in children’s spoken word recognition and segmentation ability. Developmental Review, 13, 286_ 350. Warrington, E.K., & McCarthy, R. (1983). Category speci c access dysphasia. Brain, 106, 859_ 878. Warrington, E.K., & McCarthy, R.A. (1987). Categories of knowledge: Further fractionations and an attempted integration. Brain, 110, 1273_ 1296. Warrington, E.K., & Shallice, T. (1979). Semantic access dyslexia. Brain, 102, 43_ 63. Waterson, N. (1971). Child phonology: A prosodic view. Journal of Linguistics, 7, 179_ 211. Waterson, N. (1981). A tentative developmental model of phonological representation. In T. Myers, J. Laver, & J. Anderson (Eds.), The cognitive representation of speech (pp. 323_ 333). Amsterdam: North Holland Publishing Co. Wilding, J. (1989). Developmental dyslexics do not t in boxes: Evidence from the case studies. European Journal of Cognitive Psychology, 1, 105_ 127. Young, A.W., & Ellis, H.D. (1992) Visual perception. In S.J. Segalowitz & I. Rapin (Eds.), Handbook of neuropsychology, Vol. 7, Section 10: Child neuropsychology, Part 2 (pp. 1_ 14). Amsterdam: Elsevier Science. Original manuscript received 16 January 1997 Accepted revision received 19 May 1997