Is Preschool Language Impairment a Risk Factor for Dyslexia in Adolescence?

J. Child Psychol. Psychiat. Vol. 41, No. 5, pp. 587–600, 2000 Cambridge University Press ' 2000 Association for Child Psychology and Psychiatry Printed in Great Britain. All rights reserved 0021–9630}00 $15±00­0.00 Is Preschool Language Impairment a Risk Factor for Dyslexia in Adolescence ? Margaret Snowling University of York, U.K. D. V. M. Bishop University of Oxford, U.K. Susan E. Stothard University of Newcastle upon Tyne, U.K. The literacy skills of 56 school leavers from the Bishop and Edmundson (1987) cohort of preschoolers with specific language impairment (SLI) were assessed at 15 years. The SLI group performed worse on tests of reading, spelling, and reading comprehension than agematched controls and the literacy outcomes were particularly poor for those with Performance IQ less than 100. The rate of specific reading retardation in the SLI group had increased between the ages of 8" and 15 years and there had been a substantial drop in # reading accuracy, relative to age. However, over 35 % had reading skills within the normal range and those who had had isolated impairments of expressive phonology had a particularly good outcome. Our findings highlight the limitations of discrepancy definitions of dyslexia that do not take account of the changing demands of reading over time. We argue that children’s phonological difficulties place them at risk of literacy failure at the outset of reading and that later, impairments of other language skills compromise development to adult levels of fluency. Keywords : Language impairment, SLI, dyslexia, reading disability, at-risk, phonological skills. Abbreviations : BR : backward readers ; FSIQ : Full Scale IQ ; MLU : mean length of utterance ; PI : phonological impairment ; PIQ : Performance IQ ; SLI : specific language impairment ; SRR : specifically reading retarded (A : on accuracy ; C : on comprehension) ; VIQ : Verbal IQ ; WORD : Wechsler Objective Reading Dimensions. including vocabulary and grammatical expression, as well as phonological deficits (Gallagher, Frith, & Snowling, 2000 ; Scarborough, 1990) and, from the perspective of language disability, studies of children with speech-language impairments frequently report a high incidence of reading difficulties (Aram, Ekelman, & Nation, 1984 ; Bird, Bishop, & Freeman, 1995 ; Dodd, 1995 ; Felsenfeld, Broen, & McGue, 1992 ; Larrivee & Catts, 1999 ; Snowling & Stackhouse, 1983 ; Stackhouse & Snowling, 1992 ; Stackhouse & Wells, 1997 ; Stark et al. 1984 ; Tallal, Allard, Miller, & Curtiss, 1997). Moreover, children who have problems in both oral language and phonological processing are at the greatest risk of failure (Catts, Fey, Zhang, & Tomblin, 1999). Such findings have led to widespread acceptance of the view that dyslexia and specific language impairment (SLI) may simply represent different manifestations of the same underlying disorder (Tallal et al. 1997). Studies of normally developing children have demonstrated that learning to read is strongly related to early language skills, and in particular to phonological processing abilities (Goswami & Bryant, 1990 ; Share, 1995). Links between language development and literacy are also evident in retrospective studies of children with diagnoses of developmental dyslexia" and it is well established from epidemiological studies that delays and difficulties in language development are more common in children with dyslexia than in control samples (Kinsbourne & Warrington, 1963 ; Naidoo, 1972 ; Rutter, Tizard, & Whitmore, 1970). Further evidence for a developmental link between language and literacy comes from studies of the precursors of dyslexia in the preschool years. These studies point to delays in the acquisition of oral language skills Requests for reprints to : M. J. Snowling, Department of Psychology, University of York, Heslington, York YO1 5DD, U.K. (E-mail : M. Snowling!psych.york.ac.uk). " For simplicity, we use the term ‘‘ developmental dyslexia ’’ to designate unexpected difficulties in learning to read, regardless of the terminology adopted by the original authors. We regard this as a synonym for ‘‘ specific reading retardation ’’. Dyslexia as a Less Severe Form of SLI The simplest way of conceptualising the continuity between dyslexia and language impairment is the severity hypothesis, which maintains that children with SLI and those with dyslexia have qualitatively similar impair587 588 M. SNOWLING et al. ments, differing only in degree. Severely affected children will have frank difficulties with spoken language, which will attract clinical concern and a diagnosis of SLI in the preschool period. Those with milder impairments may have some early language delay but are less likely to be seen in need of any preschool intervention : the extent of their underlying problems will only become apparent when they are exposed to literacy instruction in school. In essence, the reading problems observed in SLI children have the same underlying cause as those seen in dyslexia (in core phonological skills) ; the two disorders differ only in the severity of the language difficulties experienced outside of the reading module ; these are ‘‘ exposed ’’ in SLI and ‘‘ hidden ’’ in dyslexia. Qualitative Differences between Dyslexia and SLI ? The severity hypothesis may be too simplistic because it assumes that the reading module develops in isolation and it ignores the impact of different linguistic profiles on learning to read. Dyslexic children have well-documented deficits in phonological awareness. Children with SLI, on the other hand, typically have primary oral language deficits that extend beyond phonology, especially in the domain of morphosyntax (see Bishop, 1997 ; Leonard, 1997). A growing body of evidence suggests such deficits impair the development of reading comprehension and also affect specific aspects of word recognition (Catts et al., 1999 ; Nation & Snowling, 1998 b). Furthermore, SLI is not an homogenous condition. A study of children with developmental speech-language disorders by Catts (1993) suggests that ability to produce speech sounds distinctively may be less important than phonological awareness in predicting literacy outcome. Catts examined the reading and phonological processing skills of children with speech-language impairments in kindergarten, following them through into second grade. As expected, measures of kindergarten phonological awareness and rapid naming skill were stronger predictors of reading than either receptive or expressive language abilities. Importantly, however, children with speech-articulation difficulties in kindergarten did not develop reading difficulties in first grade, whilst those with more widespread language impairments did. A more recent study by Larrivee and Catts (1999) qualified this finding by showing that the severity of the expressive phonological disorder is an important factor affecting outcome. Similar variability in the early reading abilities of children with communication problems was reported by Magnusson and Naucler (1990). As a group, 6-year-old children with speech and language impairments were significantly behind controls matched for age, sex, and nonverbal IQ in their reading and spelling skills at the end of first grade. However, within matched pairs of children, it was often the language-disordered child whose reading and spelling skills were more advanced. Those languagedisordered children who were unexpectedly good readers were above the mean for their group in language comprehension and production, and they also did well on tests of phonological awareness. It is clear, therefore, that one needs to distinguish poor phonological production skills and phonological awareness as predictors of literacy outcome. Furthermore, there is evidence that nonphonological aspects of language may play an important role in determining literacy in children with SLI. In one of the largest prospective studies of children with preschool specific language impairments, Bishop and Adams (1990) found that, once nonverbal ability was controlled, the best early predictor of reading accuracy at 8" years was mean length # of utterance (MLU) at 4" years ; performance on a test of # grammatical understanding also contributed to the prediction of spelling, and expressive semantics to the prediction of reading comprehension. A similar picture emerged when predicting subsequent reading skill from the 5"-year-old data, although at this age a phonological # measure, namely the percentage of consonants pronounced correctly in elicited speech, also made an independent contribution to outcome. Altogether, 12 children were classified as having isolated expressive phonological problems at 4 years (Bishop & Edmundson, 1987). These children did particularly well in reading and spelling, and all but one of the six who had a persisting impairment of expressive phonology at 5" had normal # literacy skills at 8" years. # This good outcome for reading was also true more generally of children whose spoken language difficulties had resolved by 5" years. Such children showed normal # reading accuracy, reading comprehension, and spelling at 8" years, and also performed within normal limits on tests # of nonword reading and spelling. In contrast, children who still exhibited specific language impairments at 5" # had reading problems at 8" years in the context of # widespread verbal difficulties. Interestingly, when differences in nonverbal ability were controlled, the persistent SLI group did not differ from controls in terms of reading accuracy. However, there was a relatively high proportion of children (12 % of the whole sample) whose reading comprehension scores were poor in relation to Performance IQ. Overall, the clinical picture in these children bore a closer resemblance to the ‘‘ poor comprehenders ’’ studied by Oakhill and colleagues (see Oakhill, 1994), rather than to classic dyslexics. Findings such as these suggest a hypothesis of qualitative difference between SLI and dyslexia : on this view, both conditions are associated with literacy problems, but the underlying mechanisms are rather different. Retaining a distinction between processes inside and outside of the reading module, it can be argued that poor phonological skills are the principal cause of poor decoding skills in dyslexic children, who nonetheless make relatively good use of their intact oral language skills to use context to comprehend text (Nation & Snowling, 1998a). In SLI, however, limitations of phonological awareness may be implicated in some but not all cases. For these children, other language problems, particularly weak vocabulary and syntax, influence literacy development both by restricting their ability to use linguistic context in decoding text, and by affecting reading comprehension (Gough & Tunmer, 1986). The Importance of Diagnostic Criteria Our discussion so far has proceeded as if there is clearcut agreement on diagnostic criteria, but this is, of course, far from being the case. The extent to which dyslexia and SLI fall on a continuum will inevitably depend on the diagnostic criteria we adopt. Traditionally, dyslexia has been defined in terms of reading ability that is unexpectedly poor in relation to the child’s intellectual ability. Some researchers, such as Snowling (1991), have adopted a rigorous definition that is compatible with recommendations of DSM-IV (American Psychiatric LITERACY OUTCOMES Association, 1994), requiring a significant discrepancy between Full Scale IQ (FSIQ) and reading ability. Such children will, by definition, often have better language skills than children with SLI. Indeed, the majority of dyslexic readers have at least age-appropriate vocabulary and general language skills. However, other researchers have, for both practical and theoretical reasons, adopted a laxer criterion. Bishop and Butterworth (1980) noted that it is much easier to find children whose reading ability is discrepant with Performance IQ (PIQ) than cases with reading ability discrepant with FSIQ. It has also been pointed out that low verbal ability may be as much a consequence as a cause of poor literacy, because the child who seldom reads will have less exposure to the complex language structure and vocabulary of written texts (Stanovich, 1994). For this reason, the case has been made that one should not attempt to adjust literacy scores for verbal ability. Many studies of specific reading retardation do rely solely on a discrepancy between literacy and PIQ as the basis of a diagnosis. Such a definition allows for the possibility of including children whose oral language skills are relatively low, and compatible with their reading level, who might resemble mild forms of SLI. In recent years, there has been growing dissatisfaction surrounding the discrepancy definition of dyslexia (Stanovich, 1994 ; Vellutino et al., 1996). Empirical studies have shown that the core deficit in dyslexia is in phonological processing, irrespective of IQ (Shaywitz, Fletcher, Holahan, & Shaywitz, 1992 ; Stanovich & Siegel, 1994) and, as a consequence, professionals are increasingly of the view that IQ should not figure in the diagnosis of reading disability at all (Frederickson & Reason, 1995). If we define reading disability on the basis of phonological processing problems, regardless of other verbal or nonverbal abilities, then the traditional divide between SLI and dyslexia starts to break down because phonological deficits are prevalent among children with SLI, often persisting when oral language impairments have resolved (Bishop, North, & Donlan, 1996 ; Stothard, Snowling, Bishop, Chipchase, & Kaplan, 1998). The Need for a Developmental Perspective Any conceptualisation of the relationship between SLI and dyslexia needs also to incorporate the cognitive view that the language skills which contribute to literacy development change over time, placing different children at risk of failure at different stages (Whitehurst & Fischel, 2000). According to a recent connectionist account, the process of reading depends on an interactive network of connections between orthographic, phonological, and semantic representations (Plaut, McClelland, Seidenberg, & Patterson, 1996). In the early stages of acquisition, attentional resources are devoted to setting up a pathway of connections between orthography and phonology. It is at this stage that children with phonological deficits are at risk. However, as development proceeds, there is a gradual increase in the involvement of a pathway linking orthographic, semantic, and phonological units. Semantic activation through this pathway is particularly important in English for the reading of irregular or exception words, e.g. yacht, choir, that are not handled well via the phonological pathway. If this model is correct, then in addition to the problems of reading comprehension that are likely concomitants of semantic difficulties (Stothard & Hulme, 1995), we would expect 589 children with weak semantic skills to encounter difficulties in word recognition, especially difficulties in reading and spelling irregular words, even though they may have made a good start in learning to read (Nation & Snowling, 1998b). In addition, children who have difficulty with the syntactic skills that contribute to sentence processing may also falter at this stage because activation of word meanings depends at least in part upon facilitation from the sentence contexts in which they appear. On the other hand, good semantic skills may ameliorate the impact of phonological deficits over time. Many adults who were diagnosed as dyslexic in childhood do learn to read accurately enough to perform within normal limits on basic word recognition tests. However, their reading often remains slow and effortful, and deficits may be apparent when they are placed under time pressure, or required to decode unfamiliar nonwords (Pennington, Van Orden, Smith, Green, & Haith, 1990). In adolescents and adults, spelling may be a more sensitive measure of literacy problems than reading (Bruck, 1992 ; Wolff, Melngailis, & Kotwica, 1996). Data from a study by Bird et al. (1995) can be interpreted within such a framework. They studied 31 boys who were initially identified as having an expressive phonological impairment at around 5 years of age. When compared with controls matched for age and nonverbal IQ, these children obtained significantly lower reading and spelling scores at 6 and 7 years. They also exhibited significant deficits on tests of nonword reading and nonword spelling, indicating that many of them were having difficulty in acquiring alphabetic competence (Frith, 1985). When a comparison was made between children with isolated phonological impairments and those with additional oral language problems, an intriguing result was obtained. In relative terms, when compared to IQ-matched controls, both groups were equally impaired on literacy. However, the group with isolated phonological problems were of above-average IQ, and in absolute terms, their reading difficulties were much less severe, with many scoring in the normal range for their age. This suggested that these children were able to use their good intelligence and oral language skills to cope with basic literacy skills at age level, despite their persisting impairment in phonological processing. If we rely solely on the discrepancy between nonverbal IQ and literacy scores, many of these children would be regarded as dyslexic. However, in most cases literacy scores were within normal limits in relation to their chronological age. Moreover, in this study, the age to which phonological problems persisted was an important factor ; those whose problems had resolved by 5" years were the ones with the # better reading outcomes. On the basis of these findings, Bishop et al. proposed ‘‘ the critical age hypothesis ’’, which helps resolve some of the discrepant results in the literature. Expressive phonological problems constitute a risk factor for poor literacy only if they persist to 5" years, # which is the age when formal reading instruction is introduced in many U.K. schools. This proposal fits well with the vast amount of evidence linking early phonological skills with reading achievement. However, it does not address the longer-term literacy outcomes for children with language impairments. The issue of long-term outcome relates to a proposal of Scarborough and Dobrich (1990), who argued that language development follows an uneven course, with plateaus as well as periods of accelerated development of 590 M. SNOWLING et al. skills. As a consequence, periods of ‘‘ illusory recovery ’’ are apparent during development when the performance of children with language impairments can be indistinguishable from that of controls. Later, when normally developing children’s skills are in the ascendancy, the language-impaired children will again exhibit problems. Within this view, a failure to find differences in literacy skills between children with preschool language impairments and controls after the first 2 or 3 years of reading instruction may reflect the fact that normal readers have only just emerged from the beginning stages of literacy development and their skills have not yet reached a level of automaticity. Later, as they become more practised and encounter more written vocabulary, the normal readers have the resources that allow them to proceed to adult fluency, but those with SLI do not. There are a number of reasons why SLI children may not become skilled readers. First, as the phonological demands imposed by the need to read longer and less familiar words increases, even subtle weaknesses in phonological processing will render reading effortful and error-prone. Second, an over-reliance on phonological reading strategies, forced by poor vocabulary and use of sentence context, may compromise the development of word recognition. Finally, poor reading might itself lead to a decrease in motivation to read, resulting in low print exposure. In short, if the illusory recovery hypothesis is correct it is reasonable to predict a widening gap between SLI and control children, with literacy skills falling further behind. Outline and Aims of the Current Study This paper reports a prospective longitudinal study of the group of children who were diagnosed as speechlanguage impaired at age 4 years by Bishop and Edmundson (1987), and whose literacy skills were investigated by Bishop and Adams (1990) at 8 years. Stothard et al. (1998) followed this group at 15 years, making a distinction between those whose language problems had resolved by age 5" years, and those who still had # significant oral language difficulties at this age. At followup in adolescence, the former group did not differ from controls on tests of vocabulary and language comprehension skills. However, they performed significantly less well on tests of phonological processing, namely nonword repetition, sentence repetition, and spoonerisms, a test of phonological awareness. Those who still had significant language difficulties at 5" years had impairments in all # aspects of spoken language functioning at 15 years, and had fallen further behind their peer group in vocabulary growth over time. The present paper examines the reading and spelling skills outcome of these young people, in relation to their earlier language and literacy status. The questions we address are : (1) On average, do adolescents with a preschool history of specific language impairment show deficits in reading accuracy, reading comprehension, and spelling ability, and, if so, are all three aspects of literacy equally impaired ? (2) How common is specific reading retardation (dyslexia) among adolescents with preschool specific language impairments ? (3) Does the rate of literacy problems show any increase between middle childhood and adolescence ? (4) What is the link between literacy problems and spoken language impairments in this population ? Method Participants Experimental group. The language-delayed group were initially studied by Bishop and Edmundson (1987). Between 1982 and 1984, Bishop and Edmundson asked professionals for referrals of 3- to 4-year-olds who had an impairment of language development that could not be attributed to low intelligence, hearing loss, physical defect, or bilingual background, and which was not associated with a recognised syndrome such as infantile autism. Eighty-eight children were recruited and 87 were assessed at the age of 5" years. On the basis of these tests, # 19 children were classified as having low nonverbal ability (the general delay group). The remaining children were designated as having a ‘‘ specific language impairment ’’ (SLI). The children were reassessed at age 5" years, by which time 30 children from # the SLI group no longer had any evidence of language impairment (resolved SLI group). The remaining 38 SLI children continued to show language difficulties and were classified as having a poor outcome (persistent SLI group). The language-impaired children were subsequently seen by Bishop and Adams (1990) and some 7 years later by our team (see Table 1 for an overview of the longitudinal aspects of the study). In 1994 we successfully traced 85 children who had taken part in the Bishop and Edmundson study. Two children had emigrated and one child had been found to have a lowfrequency hearing loss and was therefore excluded from the study. Of these, 71 children agreed to take part in the follow-up, giving a participation rate of 87 % (82 % of the original study group). For present purposes we excluded children who were classified as having a ‘‘ general delay ’’ at 4 years. The present sample consisted of 56 young people (47 males and 9 females) aged between 15 years and 16 years 5 months (mean age 15±6 years, SD ¯ 0±38), who as preschool children had a specific language impairment (SLI). Control group. Since the majority of the SLI children had been raised in inner-city communities, a cross-sectional control group was selected from five comprehensive schools in the Tyneside area of north-east England. The school intakes covered a wide range of social classes, which were considered to be representative of the children in the experimental group. Fifty-two children within the same age range as the SLI group (38 males and 14 females) took part in the study. In one school we were able to select 26 children at random from the target age group, keeping the gender ratio the same as that in the languageimpaired sample. A further 26 children volunteered and were recruited with similar constraints. One child who volunteered had to be excluded because she was the co-twin of one of the language-impaired children. None of the controls had ever received any speech therapy. There were no significant differences between the volunteers and the randomly selected group on any measure. The control group comprised 38 males and 14 females. This gender ratio (2±7 : 1) matched that of the languagedelayed group as a whole, but differed from that in the subgroup who are the focus of this paper, which was 5 : 1. Most children were from social class III (Office of Population Censuses and Surveys, 1980) and there was no significant difference between the socioeconomic status of the language-impaired and control groups. The control group was given the entire test battery to provide normative standards. Three of the controls had estimated PIQ below 75 and were classified as having a general delay. These children’s scores were excluded from the analyses reported by Stothard et al. (1998) but are retained here. Tests and Materials Each child was administered the following tests as part of a more extensive battery (see Stothard et al., 1998, for details). LITERACY OUTCOMES 591 Table 1 Details of the Longitudinal Study Conducted by Bishop and Colleagues Bishop & Edmundson (1987) Age at test N SLI Good outcome Poor outcome General delay a 85 of the sample b Not included. 4 years Bishop & Adams Stothard et al. (1990) (1998) This study 8 years 15 years 15 years 88 5" years # 87 83 71a 56 n}a n}a n}a 32 36 19 29 37 16 26 30 15 26 30 —b were traced ; 71 agreed to participate. General cognitive ability. (1) Nonverbal ability : A short-form PIQ estimate was obtained by combining scales scores from the Picture Completion and Block Design subtests of the Wechsler Intelligence Scale for Children (WISC-III, Wechsler, 1992). Split-half reliabilities for children aged 15 years are .82 and .92, respectively. (2) Verbal ability : A short-form Verbal IQ (VIQ) estimate was obtained by combining scales scores from the Vocabulary and Comprehension subtests of the Wechsler Intelligence Scale for Children (WISC-III, Wechsler, 1992). Split-half reliabilities for children aged 15 years are .91 and .80, respectively. Literacy skills. (1) Basic reading : Word recognition reading skill was assessed by the Wechsler Objective Reading Dimensions (WORD ; Rust, Golombok, & Trickey, 1993). This test measures the ability to read single words which are presented out of context. The test has a split-half reliability of .88 for children aged 15 years. (2) Spelling : The WORD Spelling test was administered to assess spelling skills. In this task, children are asked to spell a series of single words which are presented in a sentence context. The test has a split-half reliability of .91 for children aged 15 years. (3) Reading comprehension : The WORD Reading Comprehension test comprises a series of short passages which the child is required to read (either aloud or silently). At the end of each passage, the text remains visible and comprehension questions are asked. The reading comprehension score is derived from the number of questions answered correctly. The test has a split-half reliability of .82 for children aged 15 years. (4) Nonword reading : The nonword reading test comprised 10 one-syllable and 10 two-syllable nonwords (e.g., gromp, tegwop) from the Graded Nonword Reading Test (Snowling, Stothard, & McLean, 1996). In addition, five more difficult nonwords (1 two-syllable and 4 three-syllable nonwords, e.g., pragendent) were added to the test to make it suitable for children aged 15–16 years. The test was scored on live presentation, however, responses were also audiotaped and any unclear responses checked after the assessment. The children were awarded 1 point for each nonword that was read correctly. Where appropriate, alternative pronunciations were allowed as outlined in the test manual. Control data were used to compute internal reliability of this test. Coefficient α (Cronbach, 1951), which represents the average covariance amongst items, was .77. Phonological processing skills. (1) Spoonerisms : Phonological awareness skills were examined by administering a spoonerism task (Perin, 1983). The children were required to transpose the initial sounds from the beginning of two spoken words (e.g., Paddington Bear U baddington pear). There were 11 two-word items in this test. One point was awarded for each word which was transposed correctly, provided that the correct word order was preserved, giving a maximum score of 22 points. Coefficient α, estimated from control data, was .79. (2) Nonword repetition : The Children’s Nonword Repetition Test (CNRep ; Gathercole, Willis, Baddeley, & Emslie, 1994) was administered to assess the ability to repeat unfamiliar nonsense words. The test comprises 40 nonwords, 10 each of 2, 3, 4, and 5 syllables in length (e.g., ballop, loddenapish) and the test–retest reliability is .77 to .80 for children aged from 5 to 7 years. The nonwords were presented individually on a tape player and the child was required to repeat them aloud. Procedure Participants from the experimental group were tested individually in a single session, mostly in university laboratories. The session lasted approximately 2" hours with breaks and # included assessments of language, literacy and psychosocial skills (reported elsewhere). The controls were seen individually, either at school or university. Results Table 2 shows the performance of the SLI and the control groups on the three components of the WORD test at 15 years, according to the WORD (Rust et al., 1993) manual, together with Verbal and Performance IQs from a short form of the WISC-III (Wechsler, 1992). Performance on the tests of nonword reading, nonword Table 2 Mean Scores on Literacy Tests and Short-form IQ for 15year-olds with a History of SLI and Controls, Expressed as Standard Scores Test WORD Basic Reading Mean SD WORD Reading Comprehension Mean SD WORD Spelling Mean SD Short-form PIQ Mean SD Short-form VIQ Mean SD Control (N ¯ 52) SLI (N ¯ 56) 102±35 9±90 86±32 16±80 95±94 13±20 82±63 12±10 100±37 11±70 85±82 17±10 98±56 16±10 90±48 20±90 100±25 17±50 83±73 19±80 592 M. SNOWLING et al. repetition, and spoonerisms are also shown, standardised relative to the performance of the control group that excluded those children with IQs below 75 (see Stothard et al., 1988, for details). Do Adolescents with a Preschool History of SLI Show Deficits in Reading Accuracy, Reading Comprehension, and Spelling Ability ? Performance on tests of word recognition, reading comprehension, and spelling were compared by treating them as repeated measures in a two-way analysis of variance. There was one between-subjects factor, (Group : SLI vs. control) and one within-subjects factor (Task : word recognition vs. reading comprehensions vs. spelling). The ANOVA revealed a highly significant effect of Group [F(1,106) ¯ 37±1, p ! .001], and of Task [F(2,212) ¯ 13±75, p ! .001] with no significant interaction [F(2,212) ! 1]. A rough estimate of effect size is given by η# (Linton & Gallo, 1975). For the Group main effect, η# ¯ .259, indicating that around 26 % of variance in literacy scores can be accounted for by Group status. These results confirmed that, on average, children with a history of preschool specific language impairment have poor literacy skills in adolescence. However, two points need to be noted. First, 59 % have WORD Basic Reading scores within 1 SD of the average or better (the figure for controls was 92 %), and 46 % have WORD Reading Comprehension scores at this level (the figure for controls was 83 %). Second, as reported by Stothard et al. (1998), there were significant differences in nonverbal ability as well as in verbal ability between children with developmental speech-language impairments and controls. This makes it of interest to consider how many of the sample have literacy skills that are poor in relation to their general cognitive ability. How Common is Dyslexia at 15 Years amongst Children with Preschool Speech-language Impairments ? Although the utility of discrepancy definitions of dyslexia is disputed (Shaywitz et al., 1992 ; Stanovich, 1991 ; Vellutino et al., 1996), it is nevertheless of interest to know whether the differences between groups in literacy skills are totally predictable from their nonverbal abilities. Using a regression formula to identify children whose reading accuracy scores were significantly discrepant from PIQ, Bishop and Adams (1990) reported that 8 % (7}82) of the present cohort had specific reading difficulties at 8" years. A slightly higher proportion (12 %) # had reading comprehension below predicted level. Although these figures are higher than the expected rate of 3 %, it is clear that only a minority of children in this sample could be regarded as dyslexic on this criterion. Furthermore, these estimates were based on the total sample, including those with general delay. When we exclude data from children who had nonverbal IQ below 70 at 4 years of age, the figure drops to 4}68 (6 %) with discrepant reading accuracy and 4}68 (6 %) with discrepant reading comprehension (two cases were discrepant on both measures). It is the literacy problems experienced by this subgroup of children that are the focus of the present paper, those who had specific language impairment at 5 years. We first compared the SLI and control groups by assessing how far their WORD Basic Reading score fell from the regression line that predicted word recognition from the short-form PIQ, and then we repeated the procedure for WORD Reading Comprehension. We next adopted a categorical approach to see how many children from the SLI group fulfilled criteria for dyslexia, according to various definitions. For the regression analyses, we took the performance of the control group as providing a normative standard. In this group, the correlation between PIQ and WORD Basic Reading was .413 ( p ! .05), and the regression equation for predicting WORD Basic Reading was y{ ¯ 77±3 ­ (0±254¬PIQ). The RMS residual was 9±12. Using this equation, a standardised residual score was computed for each child, reflecting the distance between the child’s WORD Basic Reading score and the control regression line. Figure 1 shows the distribution of scores of the control and SLI groups relative to the control regression line#. As can be seen from the figure, the scores for the SLI group tend to fall below the control regression line. The mean standardised residual for the SLI group was ®1±55 (SD ¯ 1±55), which is significantly below the control level (mean 0, SD ¯ 1±0) on a t-test, t(55) ¯ 7±49, p ! .001. It is apparent from inspection, however, that the degree of specific reading impairment in the SLI group depends upon PIQ. Overall, PIQ exerts a stronger influence on word recognition in the SLI group (r ¯ .643, N ¯ 56) than in the control group. We divided the groups according to whether PIQ was above or below 100. Within the SLI group, there was a significant difference between the Basic Reading residuals for children with IQs above and below 100, t(54) ¯ 3±05, p ! .01, although both subgroups had significant deficits relative to controls. For the 15 children in the SLI group with PIQs of 100 or above, the mean standardised residual was ®0±58 (SD ¯ 0±88), which is significantly below the mean of 0±03 (SD ¯ 0±75) seen in controls of this PIQ level, t(33) ¯ 1±99, p ! .05. For the 41 children in the SLI group with PIQ below 100, the mean standardised residual was ®1±90 (SD ¯ 1±59), which is significantly below the control level of ®0±2 (SD ¯ 1±14), t(71) ¯ 5±76, p ! .001. Thus, the poor word recognition skills in the SLI group cannot be accounted for in terms of low nonverbal ability. The correlation between PIQ and WORD Reading Comprehension in the control group was .411 ( p ! .05) and the regression equation for predicting WORD comprehension was y{ ¯ 61±49 ­ (0±349¬PIQ). The RMS residual was 12±58. Using this equation, a second standardised residual score was computed for each child, reflecting the distance between the child’s WORD Reading Comprehension score and the control regression line. Figure 2 shows the distribution of reading comprehension scores of the control and SLI groups relative to the regression line, with the scores for the SLI group again falling below the control regression line. PIQ was correlated with reading comprehension in both groups (SLI ; r ¯ .503, N ¯ 56, Controls ; r ¯ .411, N ¯ 52). Again, the mean standardised residual for the SLI group was ®0±84 (SD ¯ 0±84) which is significantly below the control level (mean 0, SD ¯ 1.0) on a t-test, t(55) ¯ 7±45, p ! .001. However, in contrast to the data for word # It should be noted that three control and seven SLI cases had a PIQ ! 70. LITERACY OUTCOMES 593 140 130 WORD Basic Reading (std score) 120 110 100 90 80 Key: SLI 70 Control 60 50 40 50 60 70 80 90 100 Performance IQ 110 120 130 150 160 140 150 Figure 1. Regression of WORD Basic Reading on PIQ. 140 WORD Reading Comprehension (std score) 130 120 110 100 90 80 Key: 70 SLI 60 Control 50 40 50 60 70 80 90 100 Performance IQ 110 120 130 Figure 2. Regression of WORD Reading Comprehension on PIQ. recognition, there was no significant difference between the Reading Comprehension residuals for SLI children with IQs above and below 100 (t ! 1). The next issue to consider is how many of these children met criteria for dyslexia at 15 years. Here we based our criteria on those used by Bishop and Adams (1990), who used the following categories : fashion, using Reading Comprehension rather than Basic Reading. (3) BR—backward readers. Adolescents whose reading scaled score was more than 2 SD below the mean on either word recognition or reading comprehension (equivalent to a standard score below 70), and who did not meet criteria for SRR. (1) SRR : A—specifically reading retarded on accuracy. We used the control regression equation to identify any child whose Basic Reading was more than 1±96 SDs below the value predicted from short-form PIQ. (2) SRR : C—specifically reading retarded on comprehension. This was defined in an analogous The frequency of these categories at 15 years is shown in the first section of Table 3. Compared with the control children, the proportion of normal readers is dramatically reduced, χ#(1, 108) ¯ 13±27, p ! .001. In contrast to the picture at 8 years, the literacy problems of adolescents with a history of speechlanguage impairment now include a high rate of 594 M. SNOWLING et al. Table 3 Observed Frequency of Different Forms of Reading Difficulty among SLI Group at 8 and 15 Years, According to Criteria Based on PIQ and FSIQ Regression on PIQ 15-year categorisationa SRR : A SRR : C SRR : both BR : backward reader OK : normal reader a See text for definitions. b Percentage of 15-year-old Regression on FSIQ Total % %conb Total % %conb 14 3 10 0 29 25±0 5±4 17±8 0±0 51±7 5±7 3±8 0±0 5±7 84±6 20 0 2 2 32 36±0 0±0 3±5 3±5 57±0 5±8 3±8 0±0 1±9 88±5 controls in this category. Table 4 Comparison at 15 Years of Children with Isolated Phonological Impairments at 4 Years and Controls Ability VIQ PIQ Literacy skills Basic reading Spelling Reading comprehension Nonword readinga Phonological processing Nonword repetitiona Spoonerismsa a Standard PI subgroup (N ¯ 10) Control (N ¯ 10) 102±60 (11±5) 107±60 (17±2) 107±80 (13±8) 107±10 (17±4) 101±50 (6±8) 100±10 (13±0) 94±20 (9±1) 95±22 (17±6) 105±20 (10±1) 105±20 (11±9) 103±80 (11±8) 106±60 (17±4) 95±07 (12±5) 89±14 (15±1) 115±27 (12±3) 109±04 (14±7) scores derived from control sample (Stothard et al., 1998). difficulties with Basic Reading accuracy, as reflected in the figures for SRR :A, SRR :both, and BR. Furthermore, the rates of specific reading difficulty have increased from 6 to 43 % for reading accuracy and from 6 to 23 % for reading comprehension. Of particular interest was the rate of SRR among adolescents who, as 4-year-olds, showed isolated phonological impairments. Since the dominant view of dyslexia is that it is characterised by specific phonological deficits, it was important to examine the literacy outcomes of this subgroup of SLI children, who had had surprisingly good literacy skills at 8 years. It is important to note that the preschool assessment of this sample had not included any measures of phonological awareness. Tests of rime and alliteration oddity had been attempted during pilot testing, but abandoned because too many children seemed unable or unwilling to attempt them. Our early index of phonology is therefore restricted to a measure of the percentage of consonants correct on an elicited naming task that was devised to sample a wide range of consonants. Of the 12 children reported by Bishop and Edmundson (1987) to have isolated expressive phonological impairment at 4 years (PI), 10 were available to follow-up. Examination of their residual scores for the two reading measures revealed that none of the subsample had specific reading difficulties. Notwithstanding the levels of literacy achieved by the PI subgroup, it remains possible that they would show reading deficits in relation to ability-matched controls without a history of speech-language impairment. We selected 10 adolescents from the control sample who were closely matched with the PI subsample for age, PIQ, and gender and compared their performance on tests of basic reading, reading comprehension, spelling, nonword reading, and two measures of phonological processing— nonword repetition and spoonerisms (see Table 4). A series of ANOVAs indicated that the PI subgroup performed significantly less well than controls in nonword repetition [F(1,18) ¯ 13±25, p ! .01] and spoonerisms [F(1,18) ¯ 8±92, p ! .01], indicating a cognitive similarity with dyslexia. However, their literacy skills were well within the normal range and the only test on which the group difference fell close to significance was reading comprehension [F(1,18) ¯ 4±17, MSE ¯ 110±63, p ¯ .056]. Does the Rate of Literacy Problems Show Any Increase between Middle Childhood and Adolescence ? The present study revealed a higher rate of literacy problems in the SLI group compared with the results of Bishop and Adams (1990), who studied the same population at 8 years of age. A note of caution, however, is that different reading tests were used at the two points in time, and uncontrolled differences between the two crosssectional control groups may account at least in part for these findings. Importantly, however, at 15 years, the pattern of reading impairment that characterised the SLI group was such that a substantial number had specific difficulty with reading accuracy. Indeed, even those young people whose language difficulties had resolved by 5" # LITERACY OUTCOMES 595 Table 5 Longitudinal Analysis of Literacy Outcome According to PIQ 8-yr PIQ ! 100 Short-form PIQ Reading accuracy Reading comprehension Spelling F (1,53) 8-yr PIQ & 100 8-yr SS 15-yr SS 8-yr SS 15-yr SS IQ Time IQ¬Time 83±43 (12±2) 90±90 (12±1) 84±77 (14±8) 83±23 (12±8) 80±30 (16±1) 76±93 (15±1) 77±50 (11±0) 76±10 (14±7) 109±56 (8±2) 105±24 (14±2) 102±36 (13±0) 99±68 (18±6) 103±64 (18±8) 97±32 (11±3) 88±76 (10±7) 96±76 (12±2) 57±84*** 4±52*** 25±59*** 95±30*** 7±27** 22±24*** 49±83*** 4±59* 25±32*** 11±43*** 2±10 !1 SS ¯ standard score. * p ! .05 ; ** p ! .01 ; *** p ! .001. years and for whom literacy outcome at 8" years had been # good, were impaired at 15 years in reading accuracy as well as in comprehension relative to controls (Stothard et al., 1998). Moreover, whereas literacy levels at 8 years were generally in line with PIQ, regardless of IQ level, it will be recalled from the above that the literacy scores of children with SLI who had a PIQ less than 100 were particularly likely to be below levels predicted from PIQ. A longitudinal analysis of literacy skills with the sample subdivided according to PIQ at 8 years of age is shown in Table 5. The results are complicated by a small but significant drop in PIQ between the two times of testing (it should be noted that the WISC-R was given by Bishop & Adams, 1990, and the WISC-III for this study ; see Stothard et al., 1998, for more details). However, relative to this, there was a more substantial decline in reading skills between 8 and 15 years. For reading accuracy, this effect was especially marked in those with PIQ below 100, as shown by the significant interaction between IQ level and time of test. For reading comprehension, the effect again interacted with IQ level but, in this case, it was especially marked in those with PIQ above 100, whose performance fell from the average level to just within normal limits. Spelling also declined with age, but the effect was comparatively small and did not differ in relation to IQ level. What is the Link between Language and Literacy Problems in This Population ? The results we have presented so far suggest that a higher than expected proportion of adolescents with a history of speech and language impairment reach schoolleaving age with significant literacy problems, and their difficulties cannot be fully accounted for by generally poor levels of intellectual ability. However, since in the normal population the relationship between reading skills and verbal ability is stronger than that with nonverbal ability, it could reasonably be argued that the prevalence of reading difficulty in the SLI population is entirely as expected. We therefore assessed the relationship between verbal skills and literacy abilities in this group and went on to recalculate the rate of specific reading difficulties as predicted from verbal as well as nonverbal ability. The correlation between VIQ and WORD Basic Reading in the control group was .685 and in the SLI group it was .727. The corresponding correlations for Reading Comprehension were .749 and .865. We proceeded to estimate the proportion of each group who had specific reading difficulties when verbal as well as nonverbal ability was taken into account. The regression equation for predicting WORD Basic Reading from VIQ and PIQ was y{ ¯ 64±24­(.398¬VIQ)®(.019¬PIQ) ; the RMS residual was 7±37. For Reading Comprehension we used the equation y{ ¯ 40±82­(.632¬VIQ) ®(.084¬PIQ), RMS residual ¯ 9±18. Using this method, 20 of the SLI sample (36 %) could be classified as SRR : A and a further 2 (3±5 %) were classified as SRR : both. In addition, two (3±5 %) were backward readers who did not qualify as retarded in either reading accuracy or comprehension. Using the same criteria, three of the control group were classified as SRR : A (5±8 %) and two as SRR : C (3±8 %), and one was a backward reader (1±9 %). Thus, even given the strong relationship between verbal ability and literacy skill in the SLI sample, a significant number of the adolescents with a history of SLI could be classified as dyslexic on the basis of having poor word recognition skills in relation to FSIQ. However, poor reading comprehension was well accounted for by IQ. Despite the strong relationship between VIQ (as measured here by vocabulary and comprehension) and reading ability, it remains a theoretical possibility that individual differences in phonological processing skill explain significant variance in reading skills within the SLI group. In short, it might be that core phonological deficits among SLI children mediate the association between their poor verbal ability and poor reading accuracy, just as they do in dyslexia. With the proviso that poor phonological skills at 15 years could at least in part be a consequence of poor reading, we went on to assess the concurrent relationship between phonological processing and reading skills in the SLI and control groups. To derive a measure of phonological processing skill, the raw scores of each participant on the spoonerisms and nonword repetition tests were first standardised against the mean of the control group (N ¯ 52) and a composite was derived by taking the average of these for each individual. In the same way, a composite reading measure was derived by taking the mean of the standard scores for WORD Basic Reading and Reading Comprehension. The regression of reading on phonological skills for the control and SLI groups is shown in Figure 3. The figure shows clearly that the relationship between the two skills is closely similar in the two groups (R# ¯ .648 for controls ; R# ¯ .675 for SLI). However, individuals from the SLI group fall towards the lower end of the curve, 596 M. SNOWLING et al. 130 Key: SLI 120 Control 110 Reading skill 100 90 80 70 60 50 0 20 40 60 Phonological skill 80 100 120 Figure 3. Regression of reading on phonological skills. where their poor phonological skills are linked with poor literacy skill. These findings are compatible with the hypothesis that phonological deficits are strongly related to reading difficulties but since the two sets of measures were taken at the same point in time, the current data set is unable to address the direction of causality. Discussion Consistent with the results of many other studies of children with an early history of speech and language impairment, we found that the cohort of SLI children first studied at preschool age by Bishop and Edmundson (1987) did less well on tests of reading, spelling, and reading comprehension at 15 years than age-matched controls and, in general, their poor literacy was strongly associated with poor phonological skills. Importantly, there had been a substantial drop in reading accuracy between 8" and 15 years and a corresponding increase in # the rate of specific reading retardation. In general, the qualitative nature of the reading problems experienced by the SLI group had changed to encompass wordlevel decoding as well as problems of reading comprehension. The Relationship between SLI and Dyslexia In the introduction, we posed two alternative hypotheses about the relationship between SLI and dyslexia. According to the severity hypothesis, the two conditions are merely different manifestations of the same disorder, with the SLI group experiencing more severe language difficulties. Our findings lead us to reject this hypothesis. While dyslexic children typically have problems with the development of decoding skills from the outset of learning to read, the tendency observed in the present sample of SLI children was for basic decoding skills to develop normally in the early years, with a relative decline in word recognition skills subsequently. For dyslexic children with good language skills, compensation for the reading impairment is usually possible (Frith, 1985 ; Snowling, 1987) but spelling difficulties are a residual sign of impairment (Bruck, 1992 ; Wolff et al., 1996). The developmental trajectory observed among the SLI children was quite different ; reading difficulties became more marked with increasing age, and spelling levels, though also declining, did so less than reading levels. Furthermore, SLI children with PIQs above 100 had spelling levels that were average for their age. Taken together, these findings provide unequivocal evidence that the developmental course of the two disorders is different. This evidence argues against the hypothesis that dyslexia is a mild form of SLI, emerging when the language difficulties that cause concern in the preschool years have resolved. However, despite the apparent developmental difference between SLI children and children defined as dyslexic according to discrepancy or exclusionary criteria, it remains a possibility that the core cognitive deficit in both conditions is in phonological processing. Indeed, a strong relationship between phonological skills and reading development was observed in the SLI and control LITERACY OUTCOMES groups in the present study, with the SLI group showing correlated weaknesses in both domains, as is the case in dyslexia (Stanovich & Siegel, 1994). It can be inferred, therefore, that qualitative differences between the disorders must turn on the language processes outside of the reading domain. In an interactive model of reading development (Plaut et al., 1996 ; Snowling, 1998), individual differences in reading profiles come about as phonological skills interact with semantic and syntactic resources that ‘‘ bootstrap ’’ word recognition. To the extent that SLI children have poor semantic and syntactic skills, their reading development will be compromised, especially after the basic alphabetic principle has been acquired (Whitehurst & Fischel, 2000). Outcome for Isolated Phonological Impairments Of particular interest were the literacy outcomes of adolescents who had isolated phonological impairments at 4 years. If poor phonology is at the core of dyslexia, then these individuals might be at a heightened risk of specific reading difficulties. Contrary to this prediction, their reading outcomes were normal. Importantly, however, they had significant difficulty with phonological processing tasks and their nonword reading skills were relatively weak though not significantly impaired relative to 15-year-old controls matched for nonverbal IQ. The cognitive similarity between these children and those classically defined as dyslexic is striking. With this in mind, it is interesting to note that, in a study of children at high risk of dyslexia because they had a first-degree affected relative, Gallagher et al. (2000) reported that 9}63 fulfilled criteria for specific speech disorder at 45 months, defined as scoring within the normal range for language but below it for articulation skills. If the group of children with isolated phonological impairments do show a mild form of dyslexia, then it falls to us to explain how they have avoided literacy failure. In addition to the possibility that these children were well placed to compensate for their phonological difficulties by virtue of their intact language skills, it is interesting to speculate that speech therapy intervention played a protective role in circumventing literacy problems. Vellutino et al. (1996) have shown that many children who are assessed as being at risk of reading difficulties at the end of their first year in school show a very good response to intervention, such that their reading skills are brought in line with those of their peers after just one semester of tutoring. If the subgroup of phonologically impaired children had received appropriate teaching to promote phonological skills, then this may have allowed them to avoid the delays in reading development often seen in such children. A Developmental Perspective on Reading in Children with Language Impairments As predicted by an interactive model of reading development, the literacy outcomes for the SLI group as a whole was rather poor, perhaps because they had language difficulties outside of the phonological domain that prevented compensation. The relative decline in word recognition skills between the ages of 8 and 15 years was especially true for those whose PIQ was less than 100. 597 For those whose PIQ was greater than 100, the decline was seen more markedly in reading comprehension, where levels fell from average to just within the normal range. These findings run contrary to Bishop and Adams’ (1990) critical age hypothesis and are more in line with Scarborough and Dobrich’s (1990) proposal of illusory recovery. Interestingly, whereas at 8" years, reading # comprehension was more likely to be impaired than reading accuracy, at 15 years, problems of reading accuracy were common and none of the SLI group were classified as having a specific reading comprehension deficit when FSIQ was taken into account. A qualification is that, in the 8-year-old test phase, Bishop and Adams used a reading test that requires children to answer comprehension questions without looking back at the text. The present study used a test in which the text remains visible. These differences could account for the lower levels of comprehension problems identified at 15 years. Nonetheless, based on our present findings, we hypothesise that in middle to late childhood, many SLI children have difficulty in extending a basic network of connections between orthography and phonology as a consequence of ‘‘ core ’’ phonological difficulties. When this is the case, word recognition may continue to develop if it is possible for the child to establish word-specific connections between orthography and phonology via semantics. However, as Plaut (Plaut et al., 1996) has proposed, this pathway relies upon intact semantic knowledge, as well as the ability to use sentence context in a predictive manner. It follows that this system will not function well for SLI children with vocabulary deficits and more pervasive language difficulties. The literacy outcomes of the SLI group offered preliminary evidence that high PIQ may act as a protective factor against the development of single-word reading difficulties. In cognitive terms, it is difficult to see what mechanism could account for the relationship between nonverbal ability and reading skill. However, since vocabulary and comprehension are strongly correlated with PIQ in the present SLI sample, this leads us to speculate that it is these language resources that allow for the better development of word recognition skills. Children with poorer PIQ generally do not have the semantic and syntactic skills required to use context effectively ; if they also have poor phonological resources they are unable to decode print proficiently. In line with this conjecture, SLI children with pervasive language difficulties leave school with low levels of literacy (Stothard et al., 1998). Diagnostic Criteria It was noted in the introduction that the relationship between dyslexia and SLI turns on the diagnostic criteria used. The findings that we have discussed so far lead us to believe that, although children with preschool phonological impairments in the context of only mildly affected language processes bear a surface similarity to dyslexic children in their reading behaviour, such behaviour is not typical of the group of adolescents with a history of SLI. Yet, the prevalence of specific reading difficulties among the SLI group was much higher than expected and had increased from 6 % to 43 % (for reading accuracy) between the ages of 8" and 15 years. If the term dyslexia # 598 M. SNOWLING et al. were to be taken as synonymous with specific reading difficulties, then the overlap between the groups would be more considerable than the cognitive analysis of their difficulties suggests. The use of the two terms interchangeably is therefore misleading. Moreover, from a methodological point of view, studies that recruit older samples of dyslexic children using the discrepancy definition run the risk of including individuals with more pervasive language disorders. Taken together, these findings highlight limitations of operational definitions of reading disorder that fail to take account of the changing cognitive demands of learning to read. A proper appreciation of the relationship between SLI and dyslexia requires an elaborate model of reading acquisition that details how, and at what point in development, different language resources are required. Whitehurst and Lonigan (1998) differentiate two domains of emergent literacy ; ‘‘ inside-out ’’ and ‘‘ outside-in ’’ skills. ‘‘ Inside-out ’’ refers to those skills such as phonological awareness, letter knowledge, and grapheme-phoneme abilities, which are required to gain competence in basic decoding. Variations in these skills predict first grade reading. ‘‘ Outside-in ’’ skills encompass word knowledge, grammatical awareness, and familiarity with narrative structures, and are related to children’s receptive and expressive language abilities. Interestingly, from the perspective of SLI, Whitehurst and Fischel (2000) showed that ‘‘ outside-in ’’ skills only come into play later during second grade, when understanding the meaning of text takes precedence over basic reading skills. To avoid the definitional issues surrounding the relationship between SLI and dyslexia, we suggest that a more productive way forward is to consider the child with a history of language impairment to be at risk of literacy difficulties in terms of the cognitive processes required for learning to read. Although early phonological skills are strong predictors of reading achievement (Bradley & Bryant, 1985), the present study shows that children with isolated expressive phonological impairments at 4 years can have a good literacy outcome. Children whose phonological impairments persist to the age of 5" are at # greater risk (Bird et al., 1995). However, normal reading progress cannot be assured for children who make a good start with decoding skills. As the range of written vocabulary they encounter increases, and the texts which they read become more demanding linguistically, children with a history of SLI remain at risk of reading problems because of the contribution of syntactic, semantic, and pragmatic language skills to the development of literacy. Importantly, deficits in these language abilities do not just impact on the development of reading comprehension. Good general language skills enable children to take advantage of top-down reading resources, such as the use of context, when learning to read (Nation & Snowling, 1998a ; Rego & Bryant, 1993). Children who cannot use such strategies effectively will have particular difficulty in learning how to read long words that are difficult to decode, especially if they are irregular in their spelling (Nation & Snowling, 1998b). The present findings underline the interactive nature of reading development, drawing as it does on a range of language skills. It is also important to note that literacy development has a reciprocal influence on language skills (Perfetti, Beck, Bell, & Hughes, 1987). It is particularly striking that the children in the SLI group who had resolved language impairments at 5" years were as # proficient as controls in nonword reading and spelling at 8" years, but by 15 years they were impaired not only in # word recognition but also on phonological processing tasks. It is also of note that those whose literacy skills had the poorest outcome suffered a relative decline in vocabulary skills (Stothard et al., 1998). The causal links among these different processes as reading development progresses should be the focus of future research. The findings of the present study need to be qualified with a degree of caution because it employed two different crosssectional control groups and used different reading tests at the two points in time. Acknowledgements—This research was supported by a grant 040195}Z}93}A from the Wellcome Trust to the first author. We thank the participants, their families, and their schools for their willing cooperation with the study. References American Psychiatric Association. (1994). Publication manual of the American Psychiatric Association (4th ed.) Washington, DC : Author. Aram, D. M., Ekelman, B. L., & Nation, J. E. (1984). Preschoolers with language disorders : 10 years later. Journal of Speech and Hearing Research, 27, 232–244. Bird, J., Bishop, D. V. M., & Freeman, N. (1995). Phonological awareness and literacy development in children with expressive phonological impairments. Journal of Speech and Hearing Research, 38, 446–462. Bishop, D. V. M. (1997). Uncommon understanding : development and disorders of language comprehension in children. Hove, U.K. : Psychology Press. Bishop, D. V. M., & Adams, C. (1990). A prospective study of the relationship between specific language impairment, phonological disorders and reading retardation. Journal of Child Psychology and Psychiatry, 31, 1027–1050. Bishop, D. V. M., & Butterworth, G. E. (1980). Verbal-performance discrepancies : Relationship to birth risk and specific reading retardation. Cortex, 16, 375–390. Bishop, D. V. M., & Edmundson, A. (1987). Languageimpaired 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. (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. Bradley, L., & Bryant, P. (1985). Rhyme and reason in reading and spelling. Ann Arbor, MI : University of Michigan Press. Bruck, M. (1992). Persistence of dyslexics’ phonological awareness deficits. Developmental Psychology, 28, 874–886. Catts, H. W. (1993). The relationship between speech-language impairments and reading disabilities. Journal of Speech and Hearing Research, 36, 948–958. Catts, H. W., Fey, M. E., Zhang, X., & Tomblin, J. B. (1999). Language basis of reading and reading disabilities : Evidence from a longitudinal investigation. Scientific Studies of Reading, 3, 331–362. Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16, 297–334. Dodd, B. (1995). Differential diagnosis and treatment of children with speech disorder. London : Whurr. Dunn, L. M., Dunn, L. M., Whetton, C., & Pintilie, D. (1982). British Picture Vocabulary Scale. Windsor : NFER-Nelson. Felsenfeld, S., Broen, P. A., & McGue, M. (1992). A 28-year follow-up of adults with a history of moderate phonological LITERACY OUTCOMES disorder : Linguistic and personality results. Journal of Speech and Hearing Research, 35, 1114–1125. Frederickson, N., & Reason, R. (1995). Discrepancy definitions of specific learning difficulties. Educational Psychology in Practice, 10, 195–205. Frith, U. (1985). Beneath the surface of developmental dyslexia. In K. Patterson, M. Coltheart, & J. Marshall (Eds.), Surface dyslexia : Neuropsychological and cognitive studies of phonological reading(pp. 301–330). London : Lawrence Erlbaum. Gallagher, A., Frith, U., & Snowling, M. J. (2000). Precursors of literacy delay among children at genetic risk of dyslexia.. Journal of Child Psychology and Psychiatry, 41, 203–213. Gathercole, S. E., Willis, C., Baddeley, A. D., & Emslie, H. (1994). The children’s test of nonword repetition : A test of phonological working memory. Memory, 2, 103–127. Goswami, U., & Bryant, P. (1990). Phonological skills and learning to read. Hove, U.K. : Lawrence Erlbaum. Gough, P. B., & Tunmer, W. E. (1986). Decoding, reading, and reading disability. Remedial and Special Education, 7, 6–10. Guilford, J. P., & Fruchter, B. (1973). Fundamental statistics in psychology and education (5th ed.). New York : McGraw-Hill. Kinsbourne, M., & Warrington, E. K. (1963). Developmental factors in reading and writing backwardness. British Journal of Psychology, 54, 145–156. Larrivee, L. S., & Catts, H. W. (1999). Early reading achievement in children with expressive phonological disorders. American Journal of Speech-Language Pathology, 8, 118–128. Leonard, L. B. (1997). Children with specific language impairment. Cambridge, MA : MIT Press. Linton, M., & Gallo, P. S. (1975). The practical statistician : Simplified handbook of statistics. Belmont, CA : Brooks-Cole. Magnusson, E., & Naucler, K. (1990). Reading and spelling in language-disordered children—linguistic and metalinguistic prerequisites : Report on a longitudinal study. Clinical Linguistics and Phonetics, 4, 49–61. Naidoo, S. (1972). Specific dyslexia. London : Pitman. Nation, K., & Snowling, M. (1998a). Individual differences in contextual facilitation : Evidence from dyslexia and poor reading comprehension. Child Development, 69, 996–1011. Nation, K., & Snowling, M. (1998b). Semantic processing and the development of word recognition skills : Evidence from children with reading comprehension difficulties. Journal of Memory and Language, 39, 85–101. Oakhill, J. (1994). Individual differences in children’s text comprehension. In M. A. Gernsbacher (Ed.), Handbook of psycholinguistics. San Diego, CA : Academic Press. Office of Population Censuses and Surveys. (1980). Classification of occupations. London : HMSO. Pennington, B. F., Van Orden, G. C., Smith, S. D., Green, P. A., & Haith, M. M. (1990). Phonological processing skills and deficits in adult dyslexic children. Child Development, 61, 1753–1778. Perfetti, C., Beck, I., Bell, L., & Hughes, C. (1987). Phonemic knowledge and learning to read are reciprocal : A longitudinal study of first grade children. Merrill-Palmer Quarterly, 33, 283–319. Perin, D. (1983). Phonemic segmentation and spelling. British Journal of Psychology, 74, 129–144. Plaut, D. (in press). Structure and function in the lexical system : Insights from distributed models of word reading and lexical decision. Language and Cognitive Processes, 12, 765–805. Plaut, D., McClelland, J. L., Seidenberg, M. S., & Patterson, K. E. (1996). Understanding normal and impaired reading : Computational principles in quasi-irregular domains. Psychological Review, 103, 56–115. Rego, L. L., & Bryant, P. E. (1993). The connection between phonological, syntactic and semantic skills and children’s reading and spelling. European Journal of Psychology of Education, 13, 235–246. Rust, J., Golombok, S., & Trickey, G. (1993). Wechsler Objective Reading Dimensions. Sidcup, U.K. : Psychological Corporation. 599 Rutter, M., Tizard, J., & Whitmore, K. (Eds.). (1970). Education, health and behaviour. London : Longman. Rutter, M., & Yule, W. (1975). The concept of specific reading retardation. Journal of Child Psychology and Psychiatry, 16, 181–197. Scarborough, H. S. (1990). Very early language deficits in dyslexic children. Child Development, 61, 1728–1743. Scarborough, H. S., & Dobrich, W. (1990). Development of children with early language delay. Journal of Speech and Hearing Research, 33, 70–83. Share, D. L. (1995). Phonological recoding and self-teaching : The sine qua non of reading acquisition. Cognition, 55, 151–218. Shaywitz, B. A., Fletcher, J. M., Holahan, J. M., & Shaywitz, S. E. (1992). Discrepancy compared to low achievement definitions of reading disability : Results from the longitudinal Connecticut study. Journal of Learning Disabilities, 25, 639–648. Snowling, M. (1987). Dyslexia : A cognitive developmental perspective. Oxford : Blackwell. Snowling, M. (1991). Developmental reading disorders. Journal of Child Psychology and Psychiatry, 32, 49–77. Snowling, M. J. (1998). Reading difficulties. Educational and Child Psychology, 15, 44–58. Snowling, M., & Stackhouse, J. (1983). Spelling performance of children with developmental verbal dyspraxia. Developmental Medicine and Child Neurology, 25, 430–437. Snowling, M., Stothard, S., & McLean, J. (1996). The Graded Nonword Reading Test. Bury St Edmunds, U.K. : Thames Valley Test Publishers. Stackhouse, J., & Snowling, M. (1992). Barriers to literacy development in two cases of developmental verbal dyspraxia. Cognitive Neuropsychology, 9, 273–299. Stackhouse, J., & Wells, B. (1997). How do speech and language problems affect literacy development ? In C. Hulme & M. Snowling (Eds.), Dyslexia : Biology, cognition and intervention(pp. 182–211). London : Whurr. Stanovich, K. E. (1994). Annotation : Does dyslexia exist ? Journal of Child Psychology and Psychiatry, 35, 579–595. Stanovich, K. E., & Siegel, L. S. (1994). The phenotypic performance profile of reading-disabled children : A regression-based test of the phonological core variabledifference model. Journal of Educational Psychology, 86, 24–53. Stanovich, K. E., & West, R. F. (1989). Exposure to print and orthographic processing. Reading Research Quarterly, 24, 402–433. Stark, R. E., Bernstein, L. E., Condino, R., Bender, M., Tallal, P., & Catts, H. (1984). Four-year follow-up study of language impaired children. Annals of Dyslexia, 34, 49–68. Stothard, S. E., & Hulme, C. (1995). A comparison of phonological skills in children with reading comprehension difficulties and children with decoding difficulties. Journal of Child Psychology and Psychiatry, 36, 399–408. Stothard, S. E., Snowling, M. J., Bishop, D. V. M., Chipchase, B. B., & Kaplan, C. A. (1998). Language impaired preschoolers : A follow-up into adolescence. Journal of Speech, Language, and Hearing Research, 41, 407–418. Tallal, P., Allard, L., Miller, S., & Curtiss, S. (1997). Academic outcomes of language impaired children. In C. Hulme & M. Snowling (Eds.), Dyslexia : Biology, cognition and intervention(pp. 167–181). London : Whurr. Vellutino, F. R., Scanlon, D. M., Sipay, E., Small, S., Pratt, A., Chen, R., & Denckla, M. (1996). Cognitive profiles of difficult-to-remediate and readily-remediated poor readers : Early intervention as a vehicle for distinguishing between cognitive and experiential deficits as basic causes of specific reading disability. Journal of Educational Psychology, 88, 601–638. Wechsler, D. (1992). Wechsler Intelligence Scale for ChildrenThird UK edition. London : Psychological Corporation. 600 M. SNOWLING et al. Whitehurst, G. J., & Fischel, J. E. (2000). Reading and language impairments in conditions of poverty. In D. V. M. Bishop & L. B. Leonard (Eds.), Speech and language impairments in children : Causes, characteristics, interventions and outcome. Hove, U.K. : Psychology Press. Whitehurst, G. J., & Lonigan, C. J. (1998). Child development and emergent literacy. Child Development, 69, 848–872. Wolff, P. H., Melngailis, I., & Kotwica, K. (1996). Family patterns of developmental dyslexia, part III. Spelling errors as behavioral phenotype. American Journal of Medical Genetics (Neuropsychiatric Genetics), 67, 378–386. Manuscript accepted 29 November 1999