Journal of Autism and Developmental Disorders

, Volume 44, Issue 1, pp 19–30

Evidence for Distinct Cognitive Profiles in Autism Spectrum Disorders and Specific Language Impairment

Authors

    • Neurocognitive Development Unit, School of PsychologyUniversity of Western Australia
    • Telethon Institute for Child Health Research, Centre for Child Health ResearchUniversity of Western Australia
  • Murray T. Maybery
    • Neurocognitive Development Unit, School of PsychologyUniversity of Western Australia
  • Luke Grayndler
    • C.H.I.L.D. AssociationThe Glenleighden School
  • Andrew J. O. Whitehouse
    • Neurocognitive Development Unit, School of PsychologyUniversity of Western Australia
    • Telethon Institute for Child Health Research, Centre for Child Health ResearchUniversity of Western Australia
Original Paper

DOI: 10.1007/s10803-013-1847-2

Cite this article as:
Taylor, L.J., Maybery, M.T., Grayndler, L. et al. J Autism Dev Disord (2014) 44: 19. doi:10.1007/s10803-013-1847-2

Abstract

Findings that a subgroup of children with an autism spectrum disorder (ASD) have linguistic capabilities that resemble specific language impairment (SLI) have led some authors to hypothesise that ASD and SLI have a shared aetiology. While considerable research has explored overlap in the language phenotypes of the two conditions, little research has examined possible overlap in cognitive characteristics. In this study, we explored nonword and sentence repetition performance, as well as performance on the Children’s Embedded Figures Test (CEFT) for children with ASD or SLI. As expected, ‘language impaired’ children with ASD (ALI) and children with SLI performed worse than both ‘language normal’ ASD (ALN) and typically developing (TD) children on the nonword and sentence repetition tests. Further, the SLI children performed worse than all other groups on the CEFT. This finding supports distinct cognitive profiles in ASD and SLI and may provide further evidence for distinct aetiological mechanisms in the two conditions.

Keywords

Cognitive phenotypeSpecific language impairmentAetiological overlap

Introduction

The language phenotype of autism spectrum disorders (ASDs) is complex and heterogeneous. While some individuals with ASD fail to develop functional verbal language, others develop large vocabularies and are able to string words together in complex and grammatically-correct sequences. Pragmatic difficulties tend to be pervasive in ASD, and on this basis, ASD has traditionally been considered as distinct from childhood language disorders, such as specific language impairment (SLI). SLI is recognised when children have delayed language development in the absence of intellectual, sensory, or other neurological abnormalities. The language phenotype of SLI is characterised by deficits in structural language (i.e. phonology, morphology, syntax), which contrasts with the pragmatic impairments and broader developmental difficulties that define ASD.

While ‘textbook’ cases of ASD and SLI can be readily distinguished, evidence is emerging to suggest that the boundaries between the conditions are not clear-cut. Several studies have found that a subgroup of children with ASD have difficulties with phonology, morphology, syntax and semantics that resemble SLI (Kjelgaard and Tager-Flusberg 2001; Lewis et al. 2007; Rapin et al. 2009). In addition, a significant proportion of children with SLI meet social and communication cut-offs for autism on gold-standard diagnostic measures of ASD (Leyfer et al. 2008) and phenotypic overlap remains through adolescence and early adulthood (Howlin et al. 2000; Mawhood et al. 2000). These findings have led some authors (e.g. Kjelgaard and Tager-Flusberg 2001) to hypothesise that ASD and SLI share aetiological underpinnings. In contrast, others have argued that the similarities are merely superficial and do not necessarily imply shared causation (e.g. Whitehouse et al. 2008; Williams et al. 2008).

A series of family, genetic and neurological studies have contributed to the ‘aetiological overlap’ debate. While the communication difficulty typically associated with the broader autism phenotype is restricted to the pragmatic domain, siblings of children with ASD may have language or literacy impairments and parents may also report a personal history of language or literacy difficulties (Bailey et al. 1998; Ruser et al. 2007; Tomblin et al. 2003). Further, family members of children with ASD show impaired performance on tests of nonword repetition, a putative marker for heritable language impairment (Bailey et al. 1998; Lindgren et al. 2009). Evidence from molecular genetic studies supports the findings from family studies, indicating shared genetic involvement for ASD and SLI. A locus on 7q35, Contactin Associated Protein-Like 2 (CNTNAP2) has attracted research interest as it is has been implicated in language delay in ASD and SLI (Alarcón et al. 2008; Arking et al. 2008; Bakkaloglu et al. 2008; Bradford et al. 2001; O’Brien et al. 2003; Vernes et al. 2008) as well as language development in the typical population (Whitehouse et al. 2011). Evidence from neurological studies is mixed. Structural and functional neuroimaging studies that have examined ASD and SLI in isolation have highlighted similarities, particularly in enlarged cortical volumes, reversed asymmetry and atypical lateralisation in both conditions (Herbert et al. 2003, 2004 ; Flagg et al. 2005; Hugdahl et al. 2004; Kleinhans et al. 2008; Knaus et al. 2010; Redcay and Courchesne 2008; Shafer et al. 2000). However, direct comparisons have been less conclusive, with some reporting shared neurological characteristics (Herbert et al. 2005; De Fossé et al. 2004) and others highlighting points of distinction (Verhoeven et al. 2012; Whitehouse and Bishop 2008). Thus, a rapidly accumulating body of research has focused on the ‘aetiological overlap’ debate, though findings have been mixed.

Cognitive Overlap in ASD and SLI

While substantial research effort has concentrated on overlap in the linguistic and behavioural phenotypes of ASD and SLI, relatively less attention has been paid to overlap in the cognitive domain. One purported cognitive explanation for the language impairment in SLI is a limitation in phonological short term memory capacity (Gathercole and Baddeley 1990). Measures of phonological short term memory such as Nonword (NWRep) and Sentence (SNRep) Repetition tasks in particular have garnered considerable research interest as these tasks are considered to be psycholinguistic markers for SLI (Bishop et al. 1996; Conti-Ramsden 2003; Conti-Ramsden et al. 2001). Extensive research has established that children with SLI perform worse on NWRep tasks relative to their typically developing (TD) peers (Kjelgaard and Tager-Flusberg 2001; Whitehouse et al. 2008; Conti-Ramsden et al. 2001; Ellis Weismer et al. 2000; Gathercole and Baddeley 1990). NWRep appears to be poor even in individuals whose language impairment has resolved (Bishop et al. 1996; Botting and Conti-Ramsden 2003; Conti-Ramsden et al. 2001). Importantly, parents and siblings of children with SLI also exhibit poor performance on NWRep tasks, thus the task is considered to be a heritable marker for SLI (Bishop et al. 1996).

Findings that a subgroup of children with ASD shows similar deficits on NWRep to children with SLI have stimulated interest into possible shared cognitive characteristics. In a large study of the linguistic capabilities of children with ASD, Kjelgaard and Tager-Flusberg (2001) divided the ASD group into ‘language normal’ (ALN), ‘borderline’ and ‘language impaired’ (ALI) subgroups based on their performance on the Peabody Picture Vocabulary Test-III and Clinical Evaluation of Language Fundamentals (CELF; Wiig et al. 1992) preschool, or III. The ALI subgroup had language scores below 70, or at least 2SD below the normative mean. Notably, the ALI but not the ALN subgroup performed poorly (more than 1SD below the mean) on the NWRep subtest. Kjelgaard and Tager-Flusberg (2001) subsequently argued that the observed similarity between ASD and SLI reflects substantial aetiological overlap. Results of several recent studies have supported Kjelgaard and Tager-Flusberg’s (2001) finding of poor NWRep and have also reported poor SNRep in ALI (Riches et al. 2010, 2011; Whitehouse et al. 2008).

Studies that explore whether cognitive characteristics commonly observed in ASD are found among individuals with SLI will further contribute to the ‘aetiological overlap’ debate. One particular aspect of ASD that has been the subject of extensive research has been ‘weak central coherence’, a processing bias towards local information at the expense of the global whole (Frith 1989; Frith and Happé 1994; Happé and Frith 2006). Patterns of performance consistent with weak central coherence in ASD have been observed in visual, linguistic and higher-level cognitive domains (Booth and Happé 2010; Happé 1997). The Embedded Figures Test (EFT) in particular has been used extensively to assess weak central coherence in ASD. In this task, individuals are required to find simple shapes that are embedded within complex figures and performance is thought to be facilitated by processing the geometric figures as constituent parts, rather than as wholes (Witkin 1971; Shah and Frith 1983, 1993). Several studies have found that, relative to control participants, individuals with ASD show comparable (Brian and Bryson 1996; Kaland et al. 2007; Ozonoff et al. 1991) or enhanced (Edgin and Pennington 2005; Jarrold et al. 2005; Jolliffe and Baron-Cohen 1997; Morgan et al. 2003; Pellicano et al. 2006; Shah and Frith 1983) EFT performance. Cognitive explanations for enhanced performance on the EFT are mixed, with some researchers arguing that strengths in low-level processing are accompanied by weaknesses in higher-level processing (e.g. Bertone et al. 2005; Frith and Happé 1994; Grinter et al. 2009a , b) whereas others have argued for enhanced low-level processing without weaknesses in global processing (e.g. Mottron et al. 2003; see also Mottron et al. 2006). Evidence regarding these two positions is mixed (see Grinter et al. 2010, for a review).

Central coherence has been relatively unexplored in SLI. Only one previous study has examined whether children with SLI have a tendency towards local-level processing in the visual domain. Akshoomoff et al. (2006) used the Hierarchical Forms Memory Task in which children were presented with a stimulus (e.g. a large letter ‘D’ made up of smaller letter ‘L’s) to study for 10 s. Following the study period, participants completed a 30 s distractor task, after which they were asked to reproduce the original form from memory. Unlike the TD children, the SLI children scored significantly higher when reproducing global relative to local forms. However, the effect was small and attributed to general planning or attentional difficulties rather than a specific processing bias. Akshoomoff et al. (2006) finding lies in contrast to the outcomes of studies that have implemented similar Hierarchical Forms tasks to assess local–global processing in individuals with ASD (Scherf et al. 2008; Wang et al. 2007). These studies have found that, relative to TD comparison groups, individuals with ASD present with a local-processing advantage (i.e. are faster to identify local compared to global forms). The available evidence suggests that SLI children show an opposite pattern of performance to ASD children on this measure of local–global processing, but no previous study has compared ASD and SLI children on a common test of local–global processing.

The Current Study

Possible overlap in the cognitive phenotypes of ASD and SLI can be explored by (a) identifying whether children with ALI carry cognitive characteristics of SLI, such as poor phonological short term memory and (b) determining whether children with SLI share aspects of the cognitive profile of ASD, such as weak central coherence. NWRep and SNRep tasks are considered to index phonological short term memory, and represent putative markers for SLI (Bishop et al. 1996). Further, NWRep distinguishes SLI from other childhood communication disorders (Botting and Conti-Ramsden 2003) and is a more sensitive and specific marker of SLI than language measures such as past tense marking (Conti-Ramsden et al. 2001). While the evidence regarding the cognitive processes underlying enhanced EFT performance in ASD is mixed, findings consistently indicate that individuals with ASD show intact or superior performance on this task. As such, enhanced EFT performance may represent a core cognitive feature of ASD.

A relative paucity of research has explored cognitive overlap in ASD and SLI, and no previous study has directly compared EFT performance between children with ASD and those with SLI. Yet research of this nature is important in terms of providing insight into a possible shared aetiology for the two conditions. Therefore, the present study examined possible points of cognitive convergence and divergence between ASD and SLI using NWRrep and SNRep tasks, as well as the EFT. If there is cognitive overlap in ASD and SLI, then, relative to TD children, we would expect that both the ALI and SLI groups would show poor performance on NWRep and SNRep tasks, as well as enhanced performance on the EFT. Conversely, if the cognitive profiles of ASD and SLI are distinct, then we would expect children with SLI and ALI to have similar performance on the NWRep and SNRep tasks, but that the ALI but not the SLI group would have superior performance on the EFT. Therefore, investigations of NWRep, SNRep, and EFT in ASD and in SLI may reveal important points of similarity and distinction in the cognitive phenotypes of ASD and SLI and subsequently have implications for clinical and theoretical conceptualisations of these conditions as distinct disorders.

Method

Participants

Three groups of participants were recruited (see Table 1 for participant characteristics). The first group comprised 32 children with an ASD, aged between 71 and 152 months. Each child had previously been diagnosed with an ASD according to DSM-IV criteria. We sought to verify diagnoses using the Autism Diagnostic Observation Schedule-Generic (ADOS-G; Lord et al. 2000). Fourteen of the ASD group met ADOS-G criteria for autism and a further 13 met ADOS-G criteria for autism spectrum disorder.1 Consistent with Whitehouse et al. (2007), we further subdivided the ASD group into language-normal (ALN; N = 18) and language impaired (ALI; N = 14) subgroups based on the criteria used to define language impairment in the SLI group. The ALI group was comprised of ASD children with either (a) nonverbal ability scores within 1SD of the normative mean and scores on the Test for Reception of Grammar-2 (TROG-2; Bishop 2003b) at least 1SD below the normative mean, or (b) TROG-2 score at least 1 SD below the nonverbal ability score. All ASD children had English as a first language.
Table 1

Sex distributions, and descriptive statistics for chronological age, nonverbal intelligence (WASI-Matrix Reasoning), TROG-2 and ADOS social and communication scores for the TD, ALN, ALI and SLI groups

 

TD (N = 61)

ALN (N = 18)

ALI (N = 14)

SLI (N = 19)

Male:female

33:28

13:5

12:2

15:4

Chronological age (in months)

    

 M

106.31

109.91

106.14

98.63

 SD

19.93

23.45

27.06

25.59

 Range

60–145

71–152

72–148

63–145

WASI-matrix reasoning

    

 M

9.97

11.22

10.07

11.05

 SD

3.18

2.62

2.95

2.76

TROG-2

    

 M

 

91.72

75.71

72.42

 SD

 

15.33

14.42

10.78

ADOS communication

    

 M

 

3.22

3.57

1.39

 SD

 

1.52

1.51

1.46

ADOS social interaction

    

 M

 

6.39

6.71

2.50

 SD

 

2.48

3.43

1.43

The second group consisted of 19 SLI children (aged 63–145 months) recruited from speech pathologists and specialist language schools. These schools cater for children with severe childhood language disorders deemed to require curriculum modification beyond that which could be provided in a mainstream school. We administered a series of psychometric tests to confirm the SLI diagnoses. The criteria used to define SLI for this study included: (a) nonverbal intelligence (assessed with the Wechsler Abbreviated Scale of Intelligence) within 1 SD of the normative mean; (b) TROG-2 score at least 1SD below the normative mean or TROG-2 score at least 1 SD below the nonverbal ability score; (c) no reported hearing loss, and (d) English as the first language.2

Sixty-six typically developing children (aged 60–145 months) were recruited from mainstream primary schools. Parents completed the Children’s Communication Checklist-2 (CCC-2; Bishop 2003a) and the Autism Spectrum Quotient-Child (AQ-C; Auyeung et al. 2008) to screen for communication difficulties and the broader autism phenotype. Two children were excluded as they scored above 76 on the AQ-C, which a previous study has identified as a cut-off score for clinical diagnoses of ASD (Auyeung et al. 2008). A further 3 children were excluded, two because they had low Global Communication Composite scores on the CCC-2 and one because English was the second language. Sixty-one TD children comprised the final sample (see Table 1).

Results of a one-way ANOVA indicated that there was no significant difference in age, F(3, 112) = .78, p = .51, or nonverbal reasoning ability, F(3, 112) = 1.23, p = .30, between the four groups. The ALN and ALI groups did not differ in terms of Communication, t(30) = .65, p = .52, or Reciprocal Social Interaction, t(30) = .31, p = .76, scores on the ADOS-G. While there was a significant difference between TROG-2 standard scores for the ALN and language impaired (ALI and SLI combined) groups, t(49) = 4.54, p < .001, the ALI and SLI groups did not differ significantly in TROG-2 scores, t(31) = .75, p = .46.

Psychometric Battery

Nonverbal reasoning was assessed using the Matrix Reasoning task from the Wechsler Abbreviated Scale of Intelligence (Wechsler and Chen 1999). Receptive grammatical understanding was assessed using the TROG-2 (Bishop 2003b). We derived scaled scores for each child on these tasks.

The Children’s Communication Checklist (CCC-2; Bishop 2003a) is a 70-item parent-report questionnaire designed to screen for communication difficulties in children with phrase speech. The scale is comprised of ten subscales that measure general communication difficulties (speech, syntax, semantics and coherence), pragmatic language (inappropriate initiation, stereotyped language, use of context and nonverbal communication) and behaviours commonly associated with ASD (social and interests). Standard scores with a mean of 10 can be derived for each subscale.

Phonological working memory was assessed using the Nonword and Sentence Repetition tasks from the Developmental Neuropsychological Assessment-II (NEPSY-II; (Korkman et al. 2007). In the NWRep subtest, children heard 13 pre-recorded nonsense words played through the speakers of a computer. The nonwords systematically increase in length (maximum five syllables) and complexity. The children were required to repeat the nonwords verbatim. Similarly, for the SNRep task, children listened to 17 sentences of increasing length and complexity and immediately repeated what they heard. The tasks are scored for the total number of nonwords or words pronounced correctly and the total number of errors. As per the NEPSY-II instructions, both tasks were discontinued if the child obtained a score of zero on four consecutive items. We calculated scaled scores for the NWRep. As the SNRep task from the NEPSY-II provides scaled scores only up to the age of six, we used raw scores as a measure of performance on this task.

Central coherence was assessed using the Children’s Embedded Figures Test (CEFT; Karp et al. 1971). The task required children to find a simple shape hidden within a more complex geometric figure. Participants were instructed to use a cut-out to help them find the hidden shape, and to let the assessor know when they had found it. If the response was incorrect, they were encouraged to keep looking. The children were allowed as many opportunities as necessary to find the simple shape and each incorrect response was recorded as an error. Participants had a maximum of 1 min to complete each of the 25 trials. Indices of performance were the latency to find the simple shape (averaged across all trials), number correct (i.e. number of trials on which the hidden shape was accurately located), and the total number of errors.

Procedure

Ethical approval for this study was obtained from the University of Western Australia Human Research Ethics Committee and the Western Australian Department of Education. Informed written consent was obtained from the parents for all children taking part in the study, and the children themselves provided either written or verbal informed consent prior to testing. Children for the TD group were recruited from government schools in the Perth metropolitan region. Children with ASD were recruited from the Western Australian Autism Biological Registry, an ongoing study of children with ASD and their families taking place at the Telethon Institute for Child Health Research in Perth. Children in the SLI group were recruited through advertisements placed in newsletters for the specialist language schools and through speech pathologists who distributed information packages to their clients.

Each child in the ASD and SLI groups participated in two testing sessions in which they completed the WASI-Matrix Reasoning, TROG-2, ADOS-G, NWRep, SNREP, CEFT and six tasks unrelated to those reported in this paper. Each session lasted approximately 1 h and the two sessions were conducted 1–2 weeks apart. The TD children participated in one testing session, which occurred at their school and lasted for approximately 1 h. Children were given a small reward (stationary item) for participating in the project.

Statistical Analyses

Methodological issues, such as matching clinical to control groups, have been the subject of some debate in the ASD literature (Jarrold and Brock 2004; Shaked and Yirmiya 2004). In particular, if a comparison group is selected to match an atypical (usually lower-than-average) profile, then the resulting TD group may be well-matched to the clinical group on the control measure, but there may be a “rebound” in performance for the TD group on subsequent experimental measures. Therefore, when assessed on the experimental tasks, the clinical group may appear to have a deficit, but this is an artificial finding that can be attributed to regression to the mean in the TD group. In addition, if a TD group is selected to match an atypical ability profile, then this comparison group may carry some of the characteristics of the clinical group. Any lack of group difference on experimental measures could then result from the matching rather than ‘typical’ performance in the clinical group.

Jarrold and Brock (2004) proposed an alternative analytic method designed to account for control variables while avoiding the limitations of group matching. Under their method, background variables such as chronological age, sex and nonverbal ability are used to predict the performance of an unselected group of TD children on experimental tasks (see also Brock et al. 2007; Thomas et al. 2009). Using the resulting regression equation for the TD children, predicted scores for clinical groups can then be generated and compared with observed scores, to determine whether the clinical groups have performed better or worse than expected based on the control variables. Under this regression-based approach, the TD children are not selected for their scores on the control variables. Therefore, the major advantage of this approach is that it overcomes the problem of regression to the mean that plagues matched-group designs. The use of this regression-based method had an additional advantage in the present study in that it permitted use of a single TD group in accounting for the influence of control variables on the performance of the ALN, ALI and SLI groups.

Based on Jarrold and Brock’s (2004) regression approach, we used nonverbal ability, age and sex variables from the TD children to predict scores for each aspect of NWRep, SNRep and CEFT performance.3 The resulting regression equations were used to generate predicted scores for the ALI, ALN and SLI children. The observed and predicted scores within each group were subsequently compared using a paired samples t test. A significant difference between the observed and predicted scores would indicate that the clinical group performed differently than expected based on their chronological age, sex and nonverbal ability.

Results

Nonword and Sentence Repetition Tasks

Data were missing for one ALN child on both the NWRep and SNRep tasks. Total and error scores on the NWRep and SNRep tasks were inspected for univariate outliers (scores falling more than 3SD from the group mean), but none was identified. Descriptive statistics for the observed and predicted scores and also the t test outcomes for the NWRep and SNRep tasks are presented in Tables 2 and 3 respectively. The ALI and SLI groups performed significantly worse than predicted in terms of scaled score and number of errors on the NWRep task (see Table 2). The SLI group also performed significantly worse than predicted on the raw SNRep score, and there was a trend for the ALI group to perform worse than predicted (see Table 3). There were no differences between observed and predicted scores for the ALN group for any of these measures. Further, there were no differences between observed and predicted errors on the SNRep task for any of the clinical groups. Overall, the ALI and SLI groups showed impaired NWRep performance in terms of scaled scores and number of errors. The performance of these groups was distinct from that observed in the ALN and TD groups.
Table 2

Observed and Predicted Scaled Scores for the TD, ALN, ALI and SLI groups on the NWRep Task

 

TD (N = 61)

ALN (N = 17)

ALI (N = 14)

SLI (N = 19)

Observed correct

    

 M

8.00

7.12

5.43

4.21

 SD

2.43

3.14

2.17

2.80

Predicted M

7.96

7.84

7.84

7.71

t value

−.95

−3.93

−5.37

p

n.s.

.002

.000

Observed errors

    

 M

19.44

21.00

27.79

25.26

 SD

6.96

7.67

4.56

7.71

Predicted M

19.49

19.87

19.75

20.42

t value

.608

6.28

2.49

p

n.s.

.000

.023

Table 3

Observed and Predicted Raw Scores for the TD, ALN, ALI and SLI groups on the SNRep Task

 

TD (N = 61)

ALN (N = 17)

ALI (N = 14)

SLI (N = 19)

Observed correct

    

 M

20.57

19.88

16.36

9.88

 SD

4.43

4.65

5.50

5.09

Predicted M

20.55

21.58

18.25

17.53

t value

−1.61

−1.86

−7.90

p

n.s.

.085

.000

Observed errors

    

 M

29.10

32.71

33.29

30.35

 SD

12.22

15.80

10.92

10.30

Predicted M

29.13

26.68

33.19

34.46

t value

1.73

.029

−1.45

p

n.s.

n.s.

n.s.

Children’s Embedded Figures Test

CEFT data were missing for two ALN children, one ALI child and two children with SLI. Scores on the CEFT were inspected for univariate outliers (scores falling more than 3SD from the group mean). The total correct scores for one ALN and one ALI participant were more than 3SD below the respective group mean and were subsequently removed from the analysis.4 Descriptive statistics for the observed and predicted scores and also the t test outcomes for all indices of performance on the CEFT are presented in Table 4.
Table 4

Observed and predicted scores for RT, number of trials correct and number of errors on the children’s Embedded Figures Test

 

TD (N = 61)

ALN (N = 15)

ALI (N = 12)

SLI (N = 17)

Observed RT

    

 M

12.74

10.80

15.19

18.18

 SD

4.80

3.98

5.95

5.46

Predicted M

12.71

10.90

13.18

13.89

t value

.123

−1.67

4.48

p

n.s.

n.s.

.000

Observed correct

    

 M

23.49

24.20

23.42

21.82

 SD

1.35

.77

1.24

2.32

Predicted M

23.46

23.84

23.82

23.49

t value

−1.44

1.18

−3.46

p

n.s.

n.s.

.003

Observed errors

    

 M

7.66

10.25

10.92

19.76

 SD

7.57

7.90

11.60

19.88

Predicted M

7.69

5.88

9.63

10.15

t value

−2.69

−.44

2.16

p

.017

n.s.

.046

Significant differences between observed and predicted scores for RT, total number of trials correct and number of errors emerged for only the SLI group (see Table 4). For RT and number of errors, the observed scores were significantly higher than the predicted scores. Conversely, the observed number correct for the SLI group was significantly lower than predicted. There were no significant differences between the actual and predicted scores for either the RT or the number correct variable for each of the ALN and ALI groups. Unexpectedly, there was a significant difference between the observed and predicted number of errors for the ALN group, with actual errors being higher than predicted. Therefore, while the ASD children did not show enhanced performance on the CEFT relative to the TD children on any of the measures, the SLI children exhibited a pattern of performance that was distinct from the ALN and ALI groups and worse than the TD children on both indices of CEFT performance.

Discussion

Findings that a subgroup of children with ASD has a linguistic profile reminiscent of SLI have fuelled interest into possible aetiological overlap between the two disorders (Kjelgaard and Tager-Flusberg 2001). Few studies have explored cognitive overlap in ASD and SLI, yet research of this nature would make a valuable contribution to the aetiological overlap debate. Results of the present study confirmed previous findings that children with ALI and those with SLI have similarly poor performance on the NWRep task. However, a novel finding was that, relative to the performance of the TD group, the children with SLI were slower and less accurate to find hidden shapes in the CEFT. In contrast, the ALI children completed the CEFT at levels consistent with TD group performance. This is the first study to report specific differences in the cognitive profiles of ASD and SLI.

Evidence supporting NWRep deficits in ALI is compelling and results of this study are consistent with several previous studies in the area (Kjelgaard and Tager-Flusberg 2001; Riches et al. 2011; Whitehouse et al. 2008; Williams et al. 2013). In more recent studies that have compared individuals with ALI and those with SLI on NWRep tasks, results consistently show differences in the error patterns between these two groups. As a consequence of the stopping rule of the NEPSY-II NWRep and SNRep tasks, we were unable to directly compare the error patterns in the ALI and SLI children. However, previous studies have found that when compared to individuals with ALI, those with SLI make more errors on long (four syllable) relative to short (two syllable) nonwords (Riches et al. 2011; Whitehouse et al. 2008) and are more likely to change the syntactic structure of the sentences (Riches et al. 2010). These findings may indicate that individuals with SLI have greater phonological short term memory limitations and more syntactic difficulties than individuals with ALI, which in turn may suggest that the structural language impairment in ALI and in SLI may have distinct cognitive or linguistic underpinnings.

The poor performance of the children with SLI on the CEFT was a unique finding. While the ASD children in our sample did not show enhanced performance on the CEFT, which would be expected based on the weak central coherence account (Frith 1989; Frith and Happé 1994; Shah and Frith 1983, 1993), the results did reveal a point of distinction in the cognitive profiles of ASD and SLI. There is inconsistency in the results from studies of CEFT performance in ASD, with some finding superior performance relative to TD controls (Edgin and Pennington 2005; Jarrold et al. 2005; Jolliffe and Baron-Cohen 1997; Morgan et al. 2003; Pellicano et al. 2006; Shah and Frith 1983), and others finding no performance difference (Brian and Bryson 1996; Ozonoff et al. 1991). The results of the current investigation are consistent with the latter studies. In explaining the non-superior EFT performance of individuals with ASD, Brian and Bryson (1996) proposed that superior EFT performance, when observed, could reflect ‘intact’ disembedding in individuals with otherwise low nonverbal ability. In support of Brian and Bryson’s (1996) claim, the ASD children in our study had average, if not slightly above average nonverbal ability. In addition, we controlled for nonverbal ability inasmuch as we used WASI-Matrix Reasoning scores to derive predicted scores for the clinical groups, which may explain the lack of difference between observed and predicted scores in ASD. However, recent evidence from studies of individuals in the general population who have high levels of autistic-like traits (measured using the Autism Spectrum Quotient [AQ]; Baron-Cohen et al. 2001) is discordant with Brian and Bryson’s (1996) claim. For example, several studies (Almeida et al. 2010a, b; Grinter et al. 2009a, b, Russell-Smith et al. 2010) have found superior EFT performance in individuals with high AQ scores, compared to those with low AQ scores, despite the two groups being matched on their relatively high levels of nonverbal ability. Importantly, it is possible that the complexity of the EFT has influenced the inconsistent research findings. Studies that have implemented simpler search tasks have found that scores on the simpler task are correlated more strongly with AQ scores than scores on the EFT (Almeida et al. in press). Therefore, administering simpler disembedding tasks to individuals with ASD may yield more consistent findings.

Finding that children with SLI had poor performance on the CEFT is consistent with results reported by Akshoomoff et al. (2006), who found that children with SLI performed worse than TD children on a Hierarchical Figures Task. The poor CEFT performance in children with SLI could be indicative of a broader difficulty in visuospatial processing, despite having normal nonverbal ability (Hick et al. 2005; Kamhi et al. 1988). Evidence regarding visuospatial skills in SLI is mixed. On the one hand, Marton (2008) reported selective deficits for children with SLI on tests of visuospatial ability, including space visualisation, position in space and design copying. On the other, several studies have found that children with SLI have visuospatial skills that are comparable to their chronological age-matched peers (Archibald and Gathercole 2006; Henry et al. 2012). A further explanation for the slow latency to find the hidden shapes in the CEFT for children with SLI could be that these children have a generally slowed processing rate. Certainly, studies that have explored information processing limitations in children with SLI have found that, relative to chronological and nonverbal-age matched TD children, children with SLI have slower RTs on non-linguistic tasks including mental rotation and simple visual search tasks (Johnston and Ellis Weismer 1983; Miller et al. 2001; Windsor and Hwang 1999). The alternative explanations advanced by Akshoomoff et al. (2006), such as children with SLI having planning or attention difficulties, could also apply to our results. Results reported by Marton (2008) provide support for a possible link between poor attention and visuospatial difficulties in SLI, as those with ‘poor attention’ performed worse than those with ‘good attention’ on visuospatial tasks. Regardless of the explanation for poor CEFT performance in SLI, the profile is distinct to that observed in the ASD children.

Overall, finding that children with SLI and those with ALI share NWRep and SNRep difficulties suggests that there are some similarities in the cognitive deficits associated with language impairment in the two groups. While it is possible that the poor NWRep and SNRep in ALI and SLI result from different cognitive or linguistic underpinnings, the poor performance of the SLI group on the CEFT was not observed in children with ALI. These findings suggest that there may be a unique aspect of the SLI cognitive phenotype that is associated with either poor visuospatial skills, slowed information processing, difficulty inhibiting gestalts, or limited planning or attentional control. Further, these findings are consistent with previous studies that have reported cognitive deficits in children with SLI who have normal nonverbal ability (Hick et al. 2005). The extent to which these deficits contribute to the language impairment in SLI is unclear, though Ellis Weismer and Evans (2002) proposed that the language impairment could be secondary to information processing limitations. Children with SLI have broad processing limitations, which affect verbal abilities, such as phonological short term memory and subsequent language development. These processing limitations may also lead to the observed visuospatial deficits. Therefore, cognitive deficits, such as a processing limitation, may underlie the language impairment in SLI but not ALI, which points to distinct mechanisms operating in these two conditions. Alternatively, the language difficulties experienced by children with SLI may restrict the skill base and hence the strategies that these children are able to use when completing visuospatial tasks (Hick et al. 2005). A further possibility is that the observed visuospatial deficit in SLI is associated with, but does not directly influence the structural language impairment of this population, just as ‘weak’ central coherence in ASD is an associated feature of ASD, but does not necessarily affect the core diagnostic features of the condition. Nonetheless, it would be useful to investigate the underlying bases of the visuospatial impairments in SLI and whether they are independent of, or form part of the cognitive processes that contribute to the language impairment in these individuals.

These results can be considered in light of the debate around shared aetiology in ASD and SLI. While there is accumulating evidence supporting aetiological overlap between ASD and SLI, findings of discrepant cognitive profiles lend further support to the claim that overlap between ASD and SLI may be superficial (e.g. Williams et al. 2008). Nonetheless, a subgroup of individuals with ASD has a language impairment that resembles SLI, which requires consideration. Whitehouse et al. (2008) proposed each aspect of the ASD triad is associated with a distinct genetic influence, which may disrupt processing in distinct neurological regions. Some children may inherit only one aspect of the cognitive or linguistic phenotype of ASD, others the full constellation of ASD traits. Still others may inherit the full constellation of ASD characteristics, which may affect subsequent language development and result in the ‘double hit’ of ASD and comorbid SLI. The observed linguistic difficulties in ALI and SLI may arise through a shared phonological short term memory deficit, but it is equally possible that the observed NWRep deficit in ALI is consequent to the broader developmental difficulties experienced by this population. Nonetheless, while the causal mechanisms in ASD and SLI are subject to considerable debate, evidence is mounting to suggest that distinct aetiological pathways may underlie the language impairment in each condition.

Concluding Remarks

Strengths of the current study include the data analytic procedure, which allowed us to control for background variables such as nonverbal intelligence, and subsequently overcome problems associated with matching in studies of ASD and SLI. However, it is also important to acknowledge some methodological limitations. It may have been useful to include a standardised measure of expressive language, which would have allowed us to examine any potential differential contributions of expressive and receptive language to performance on the CEFT task. In addition, we did not measure articulation or oromotor skill in the ASD and SLI children. Therefore, the NWRep deficits could be explained by oromotor deficits in the ALI or SLI children, rather than a cognitive deficit. However, in a recent comparison of NWRep performance in ASD and SLI, Williams et al. (2013) included only SLI children with no articulation impairment and still found poor NWRep in these children. Therefore, the NWRep deficit in SLI may not solely be explained by concurrent articulation difficulties. Further, while investigations of the error patterns in ALI and SLI provide an important contribution to the debate about cognitive overlap, we were unable to conduct such analyses as a consequence of the stopping rule for the versions of the NWRep and SNRep tasks we administered.

Finding that ASD and SLI have distinct cognitive phenotypes provides insight into possible distinct aetiological underpinnings. Future research should continue to attempt to elucidate the specific cognitive and linguistic contributions to poor NWRep and SNRep in ALI and SLI to further inform the debate on aetiological overlap in these conditions. In addition, considering the influence of attention, processing speed and visuospatial ability in explorations of CEFT performance in SLI is likely to enhance our understanding of the cognitive profile of this condition, which in turn will further our knowledge of cognitive links between ASD and SLI. Understanding these links is likely to clarify the boundaries between the conditions and contribute to improved diagnostic and intervention practices.

Footnotes
1

Five of the ASD children did not meet autism spectrum cut-offs on the ADOS-G. For two of these children, the communication and social interaction total score was only one point below the ADOS-G ‘autism spectrum’ cut-off. All of these children had also previously been diagnosed with an ASD by a team that included a speech pathologist, clinical psychologist and paediatrician.

 
2

CCC-2 questionnaires were returned for 15 ALN, 10 ALI and 10 SLI children. Consistent with Whitehouse et al. (2008), each of these children had GCC scores below the 10th percentile. While the return rates were not high enough to warrant extensive use of the CCC-2 results, where available, the scores confirm the composition of the ALN, ALI and SLI groups.

 
3

This data analytic approach is used to control for extraneous factors such as age, gender and nonverbal ability that may potentially affect experimental performance. As the groups were selected to differ systematically on language ability and ASD traits, it is the influence of these factors that were under investigation. Therefore predictors tapping language ability or ASD traits were not introduced to the regression model.

 
4

Though outcomes of the t tests that compared the observed and predicted scores were unchanged when these scores were retained.

 

Acknowledgments

Lauren Taylor is supported by an Australian Postgraduate Award, and Andrew Whitehouse by a NHMRC Career Development Fellowship (#1004065). This Project was supported in part by an Apex Foundation Trust for Autism Research PhD Grant and the ARC Discovery Project Grant DP 120104713. We would like to thank all of the children and families for participating in this study. In addition, we would like to acknowledge the following schools for their assistance with recruitment and for providing space to conduct assessments: Connolly Primary School, Ocean Reef Primary School, Currambine Primary School, Heathridge Primary School, South Ballajura Primary School and The Quintilian School. Thanks also to the West Coast Language Development Centre, the Autism Association of Western Australia and Anna Hunt for their help with recruitment. We extend our appreciation to Doris Leung for developing the tasks and to Pamela See and Roxanne Smith for their help with the assessments. Many thanks also to the staff and students at The Glenleighden School for their support, and for welcoming us into their school.

Copyright information

© Springer Science+Business Media New York 2013