Complex syntax in autism spectrum disorders: a study of relative clauses

Introduction

Delays and deficits in language and communication are part of the defining characteristics of autism spectrum disorders (ASD), and difficulties in speech are amongst the first concerns by parents of children with this condition (Short and Schopler 1988). Language difficulties are however heterogeneous: while 15% of the population remains nonverbal at age 9 (Gotham et al. 2010), at the other end of the spectrum no residual grammatical impairment may be visible at the same age (Kelley et al. 2006). However, even when fluency is ultimately attained, a diagnosis of autism would only be given with the previous Diagnostics and Statistical Manual of Mental Disorders IV—Revised (DSM-IV-TR) (American Psychiatric Association [APA] 2000) with a history of language delay. According to the DSM-IV-TR, autistic traits without language delay during childhood would give rise to a diagnosis of Asperger syndrome. The distinction between these subgroups is no longer expressed in the recent DSM-5 (APA 2013), which instead situates autism and Asperger syndrome at different levels of severity along the same continuum, referred to as ASD. Nevertheless it is well known that the age of language acquisition impacts the level of language reached in typical development (Johnson and Newport 1989) and work on ASD also suggests that early delays in this population may affect their eventual attainment of grammatical abilities (Paul and Cohen 1984, Eigsti and Bennetto 2009). With this work, we explore whether adults with ASD within normal IQ range show differences in performance to normal age-matched controls on a task of complex syntax targeting relative clauses and if these potential differences are more pronounced in the subgroup of adults diagnosed with high-functioning autism than the one with Asperger syndrome, due to their different language development histories. In light of the recent DSM-5, we will refer to these groups as ASD with LD (language delay in childhood) and ASD without LD and, by comparing their respective levels of syntactic comprehension, will evaluate the impact of language acquisition delays in individuals with ASD and intelligence within normal range.

This study thus contributes to an on-going debate in the literature regarding the nature of syntactic impairment in ASD, as the few studies that have evaluated syntax in ASD have yielded conflicting findings (Eigsti et al. 2007). While earlier studies have claimed that once matched on mental age, ASD and typically developing (TD) controls do not differ for grammatical development (Bartak et al. 1975, Pierce and Bartolucci 1977, Tager-Flusberg et al. 1990), in more recent years new studies have argued that morphosyntactic deficits can be independent of cognitive skills (Eigsti et al. 2007, Kjelgaard and Tager-Flusberg 2001, Roberts et al. 2004, Zebib et al. 2013) and insist that specific grammatical deficits exist in ASD (Perovic et al. 2013, Terzi et al., 2012). More specifically, researchers have now shown that some subgroups with ASD present syntactic profiles reminiscent of specific language impairment (SLI) (e.g. Roberts et al. 2004, Tager-Flusberg 2006, Riches et al. 2010), and that performance on syntactic tasks does not correlate with IQ measures such as those yielded by Raven's Matrices (Zebib et al. 2013). One possible explanation for these differences in reports may be due to the fact that gross measures of emerging syntax based on language samples were generally used in earlier investigations (e.g. The Mean Length of Utterance, Brown 1973; The Index of Productive Syntax, Scarborough 1990), while more fine-grained, experimental tasks tapping into specific grammatical abilities (such as tense and wh-questions) have been applied in recent work. However, despite these recent experimental investigations of syntax in ASD, data are still lacking regarding a number of important structures characterizing complex syntax. In addition, the question remains whether or not the delay in language acquisition during the childhood of individuals with ASD of intelligence within the normal range leads to deficits that remain detectible in adulthood and which may potentially differentiate this group from adults with ASD without a history of language delay, who are reported to have normal syntactical speech (Tager-Flusberg and Joseph 2005, Slocombe et al. 2013).

Example (1) illustrates a case of subject relative clause (SR), thus called because the relative pronoun that designates the subject of the clause (the dog). Example (2) on the other hand is a case of object relative clause (OR), with the pronoun that this time designating the object of the clause (the cat).

In typical language development, ORs are acquired later than SRs (Friedmann et al. 2009, Adani 2011, Bentea and Durrleman 2013). This difference in acquisition is generally explained by the fact that SRs respect the canonical subject–verb–object word order in languages such as English and French, while ORs do not. In an OR, the object is fronted in the clause, and the word order thus becomes object–subject–verb. By assessing both types of relative clauses in individuals with ASD, we can determine if a form of impairment is already present at the level of the simpler subject relatives, or if their potential difficulty is only manifest at the more difficult OR level. Our choice of relative clauses was also motivated by the fact that they have been shown to be a useful tool for detecting syntactic difficulty in various conditions such as SLI (Friedmann and Novogrodsky 2004), mild-to-moderate hearing loss (Delage 2008) and agrammatic aphasia (Grodzinsky 1989). Because of this body of existing studies, the scores of individuals with ASD can also be compared with that of other populations, especially children with SLI, in order to validate possible convergences between the linguistic abilities within these two conditions.

To sum up, if syntax in ASD is intact, then performance of the clinical group on relative clauses should be in the same range as that of their age-matched control group, whereas the presence of deficits should yield significant differences in results between ASD and controls. In addition, if a history of language delay in childhood has consequences for grammatical abilities attained in adulthood, then individuals with ASD with LD should perform differently from those without LD, while a similar level of performance is expected for both subgroups if childhood delays are caught up. Finally, if syntax in ASD is similar to that reported for SLI on a similar task, then subject relative clauses should be intact in the population, with object relative clauses showing selective impairment.

Methods

Results

We first compared global success scores of the ASD group (i.e. including both LD vs. no LD) with the control group on the ceiling versus not ceiling distinction, as explained above. The contingency table is reported in table 1. Fisher's exact test indicates that the difference between the number of individuals reaching ceiling between groups is significant (p < 0.001). Post-hoc tests showed significant differences between ASD without LD and controls, and ASD with LD and controls (both p < 0.0001). ASD with LD did no differ from ASD without LD for the global success score (p = 0.37).

Results from the GLMM revealed a significant main effect of relative clause type (the likelihood of success at ORs being lower than SRs, z = –7.34, p < 0.0001), and the analysis revealed a main effect of language delay (z = 2.64, p = 0.008), the group without LD having a higher probability to succeed at the task than the group with LD. We also observed the presence of a significant interaction between relative clause type and diagnosis (z = –2.67, p = 0.007), showing that individuals without LD have a higher probability to succeed at SRs than those with LD while the two groups do not differ for ORs (figure 1).2

Subsequent GLMM analyses reveal that the ASD group without LD had a higher probability to succeed the SRs than the ORs (z = –3.73, p = 0.0001), while ASD participants with LD did not (z = –0.71, p = 0.47). In addition, IQ was also observed to have a significant impact on success (OR and SR confounded, score between 0 and 28) at this task for the ASD group (GLMM: z = 3.42, p < 0.001) when tested as only covariate. Similarly, analyses exploring correlations between performance and age also showed a significant positive correlation (p = 0.013).

The Spearman's rho revealed a statistically significant negative relationship between the age at which subjects produced their first words and their ability to understand subject relative clauses (r = –0.84, p = 0.017). No significant relationship was found between the age at which subjects produced their first words and their ability to understand object relative clauses (r = –0.26, p = 0.23).

Discussion

Our experiment reveals subtle difficulty with the comprehension of relative clauses by the ASD population when compared with results of age-matched controls. While the former show variability in the processing of these structures, CA controls were virtually at ceiling. The variability in performance amongst the ASD population was further seen to correlate with performance IQ (PIQ), with higher PIQ implying also better language performance. This is in contradiction to that which has been claimed in a recent study by Zebib et al. (2013), where IQ scores for 20 ASD participants (mean age 8;7, IQ range 48–108) as measured by Raven's Matrices were reported as not correlated with formal language abilities. Given that the majority of the participants had a diagnosis of ASD without LD, and that analyses had to be conducted on the group as a whole for statistical reasons, it would now be interesting to see if this correlation between PIQ and syntactic skills also holds in a group made up of only individuals with ASD with LD. We leave this for future work.

Our results indicate that individuals with ASD, regardless of their language development history, have a significantly higher probability to master SRs than ORs. This is reminiscent of what is reported for ASD by Riches et al. (2010) who also note more errors with ORs than SRs in a sentence-repetition task. This pattern of improved performance for SRs than ORs is also what is generally reported for TD children (Berman 1997, Brown 1972, Correa 1982, 1995, McKee et al. 1998, Roth 1984, Sheldon 1974, Tavakolian 1981; Friedmann et al. 2009), indicating that language in high-functioning ASD appears to be slightly delayed rather than fundamentally deviant. More specifically, ORs, emerging later in typical acquisition, yield lower scores in ASD than structures which are acquired earlier by TD children, namely SRs. Future research should seek to determine at what age TD children's scores for ORs match those of the adult ASD group. However, based on results from similar tasks in the literature, we predict that this age should be situated around 10 years (Bentea and Durrleman 2012). Regarding SRs, these are mastered early in typical development, with children age 6 years already reaching 95% accuracy (e.g. Friedmann and Novogrodsky 2004). It is therefore noteworthy that these structures are still not at ceiling in adults with ASD, and scores obtained are even below those reported for children with SLI age range 7;3–11;2 on a similar task, which is 98.5% (with a standard deviation of .05%; Friedmann and Novogrodsky 2004).

Lower scores for ORs as compared with SRs plausibly stem from a combination of two complexity factors: (1) embedding and (2) the non-canonical (object–subject–verb) word order derived by the fronting of the object. Object fronting involves a process referred to as ‘syntactic movement’. Challenges with such movement are commonly reported in language-impaired groups, and have recently been found in ASD populations (Perovic et al. 2007). Studies also suggest difficulty in ASD with embedding (e.g. Tager-Flusberg 2000, Durrleman and Zufferey 2013) while embedding appears to be spared in SLI (e.g. Fattal et al. 2011).

The reported weakness with embedding in ASD could be the source of the attested subtle difficulty for the participants with SRs, and potentially help to differentiate the language profile of ASD from that of SLI. Indeed with a similar procedure, SRs have been found to be mastered by children with SLI (Friedmann and Novogrodsky 2004), with ORs showing selective impairment. The pattern of slightly lagged performance of SRs in the population of adults with ASD has moreover been replicated for children with this condition (Durrleman and Franck 2012).

Scores obtained by the subgroup of participants with ASD without LD indicate that they perform better than individuals on the spectrum with LD, in particular for SRs, suggesting a difference between these groups in terms of language development, in line with the distinction that was previously made in the DSM-IV-R. This difference could stem from the presence or absence of observable language delays in children with ASD, an explanation which is upheld by the fact that scores for SRs were lower for adults with ASD whose first words were produced later than for those whose first words were produced earlier. Participants with ASD without LD nevertheless do not perform completely on a par with adult controls. This shows that a slight delay may be present during childhood for these individuals but go undetected by standardized language assessments. It becomes conceivable that these individuals may have struggled in school as the demands for language complexity increased. Clinicians should be aware of these subtle differences so as to accordingly implement interventions. Given that only subtle deficits show up with the task involving response accuracy, the use of a more refined measure such as response time or reading time may prove more effective at uncovering possible syntactic difficulties in this group. Also, work assessing a wider array of syntactic structures would potentially uncover additional sources of grammatical delay, although it lies outside the scope of the present study. Finally, given that the participants showed variability on age, and in light of recent findings indicating a potential delay in maturation of the white matter of the brain in ASD, including in language networks (Bakhtiari et al. 2012), links between performance and age were explored and results showed a significant positive correlation. This finding suggests that amelioration continues during a later phase in ASD as compared with that observed in TD, as suggested by Bakhtiari et al. (2012).

Conclusion

This study of relative clauses reveals that subtle grammatical difficulties remain in adult individuals with ASD within normal IQ range as compared with CA controls. Higher scores with SRs than with ORs are reminiscent of the pattern reported for typically developing children, suggesting that a language delay in the acquisition of these structures may lead to a plateau of certain abilities in adulthood. The level of performance for SRs is different to that reported in the literature for children with SLI for a similar procedure, potentially indicating a distinction between these groups. A targeted study including populations with ASD and SLI evaluated with the same experimental task is needed to address this apparent difference in syntactic profiles. The increased difficulty of ORs attested in the population with ASD can be explained in terms of the additional challenge for the syntactic computational system presented by these structures. Even in absence of a history of language delay in childhood, the results show that a slight delay may nevertheless be present and go undetected by standardized language assessments. Future research involving more refined measures may allow to further highlight subtle impairments in this population. Even though both groups of individuals with ASD have a similar global performance in the task, the study also reveals differences, with individuals with LD showing more difficulty with SRs than individuals without LD. This suggests that some distinctions in linguistic abilities exist between ASD adults with and without LD in childhood.