Autism Spectrum Disorder and Specific Language Impairment: Overlaps in Syntactic Profiles

ABSTRACT

This study investigates syntax in Autism Spectrum Disorders (ASD), its parallelism with Specific Language Impairment (SLI) and its relation to other aspects of cognition. We focus on (1) 3^rd person accusative clitic (ACC3) production, a clinical marker of SLI hypothesized to relate to WM, and (2) 1^st person accusative clitic (ACC1) production, preserved in SLI but hypothesized to be affected in ASD due to Theory-of-Mind (ToM) difficulties. Participants included 21 individuals with ASD (aged 5-16), 22 individuals with SLI (aged 5-16), age-matched and younger TD controls (N = 44). Clinical groups showed similar deficits for ACC3 and general morphosyntax. Closer analysis revealed that a subgroup of children with ASD displayed intact grammar except for ACC1, where children with SLI performed well. Better ToM scores implied better ACC1 scores in ASD. Difficulties with WM emerged for ASD and SLI and correlated only with performance on ACC3. Non-verbal reasoning was unrelated to syntactic measures.

Notes

¹ In other languages, it is not the actual production of clitics that is problematic, as reported for French, but rather clitic misplacement (see Theodorou & Grohmann 2015).

² Measured by a variety of simple and complex span tasks, including nonword repetition, serial memory, forward and backward digit span, as well as counting span.

³ The particularity of this standardized intelligence measure is that it is essentially nonverbal, including 36 items focusing on visual problem solving.

⁴ It is relevant to note that there is no difference between French from France and French from Switzerland for the clitics under investigation.

⁵ The raw score corresponds to the number of correct answers (maximum score = 36). In this task, norms are given as percentile ranges. For statistical convenience, we converted the percentile ranges into midpoint percentiles (e.g., 5th to 10th percentile is transformed into 7.5th percentile), along the lines of Zebib et al. (2013).

⁶ See Appendix B for more information regarding this bilingual variable, which did not show a relationship in our sample for any of the factors explored in this work.

⁷ This same strategy was applied throughout the elicitation probe for both ACC1 and ACC3. To illustrate how this applied to ACC3: Que fait Thomas à Marie? ‘What is Thomas doing to Mary?’. Expected response: il la pousse ‘He is pushing her.’ Inappropriate response: Va-t-en! ‘Go away!’ Experimenter: oui, regarde ça s’appelle ‘pousser’. Que fait Thomas à Marie? Comment tu pourrais dire avec ‘pousser’? ‘Yes, look, that’s called “to push.” What is Thomas doing to Mary? How could you say it with “push”?’ Expected response: il la pousse ‘He is pushing her.’ For the ASD group, 76 prompts for reformulation were necessary, and 21 of them yielded a correct production including the relevant clitic, i.e., 28%. For the SLI group, 47 prompts for reformulation occurred with 14 leading to correct responses, i.e., 30%.

⁸ We removed these phonologically simple nonwords in order to shorten the task and avoid ceiling effects on these extremely simple nonwords.

⁹ As the expressive grammar task is a standardized task, we do not have scores for a specific TD sample, but real norms.

¹⁰ This result can seem surprising in light of a diagnosis of SLI. A series of observations may be relevant here. Firstly, on the well-respected standardized tests used by professionals to diagnose diagnosed SLI, these children did indeed score below the threshold required for this diagnosis (test of Syntactic comprehension [ECOSSE, Lecocq 1996], which is the French adaptation of the Test for Reception of Grammar [TROG, Bishop 1983]). Secondly, all participants with SLI had a level of language difficulty that continued to justify their being included in remediation programs, which is where they were recruited. Thirdly, the “BILO” standardized test used here comprises items that are very frequently used by speech therapists during their language evaluations and interventions. Due to this, participants with SLI may have received a little “training” on these items. However, amongst these six participants with SLI who managed to perform well on the BILO, three nevertheless display severe phonological disorders (with SD < –2 on nonword repetition), two present difficulties on ACC3 production (with low percentages of correct production, compared to age-matched controls), and only one child does not present any other difficulties. Since he is among the oldest participants and had already benefited from a lengthy speech and language therapy, we hypothesize that he may have compensated for his initial difficulties.

¹¹ These controls do not differ from the ASD and SLI groups for age (p = .5 for both comparisons).

¹² Interestingly, this difficulty is limited to accusative clitics, since both ASD and SLI groups display good performance for production rates of third person nominatives clitics (85% and 94% respectively).

¹³ Note that the difference between ACC1 and ACC3, in favor of ACC1, has been shown in younger typically developing children, aged 6, by Tuller et al. (2011), as already explained in the introduction.

¹⁴ Forward digit span: Controls > ASD → U = 315.5, p = .022, Controls > SLI → U = 206.5, p < .001; Backward digit span: Controls > ASD → U = 234, p = .013, Controls > SLI → U = 235.5, p < .001.

¹⁵ Interestingly, no differences were found for nonword repetition (p = .1), nor for forward and backward digit span (p = .7 and p = .5), showing that these measures of working memory do not seem related to nonverbal reasoning abilities.

¹⁶ For complex sentences, they included structures like passives, while simple sentences did not express movement and thus respected canonical word order.

¹⁷ As for expressive grammar, these 14 children obtained a mean standard deviation of –2.5 on standardized assessment of grammar, and they displayed production rates of ACC3 and ACC1 of 54% and 60%.

¹⁸ This subset of participants was determined based solely on availability to participate.

¹⁹ We did not test children with SLI for ToM, given that the aim of this testing was to explain difficulties in ACC1, which were absent from SLI. However, as one reviewer points out, this would have been appropriate as we predict good performance for this group, although some studies have reported a ToM delay in SLI (Holmes 2002; Tucker 2004; Roqueta et al. 2013). It is worth noting that the ToM delay reported was detected in very young children with SLI, while studies assessing ToM in SLI subjects of a comparable age to our participants have found no difficulties with these tasks, unlike their ASD peers (Leslie & Frith 1988; Perner et al. 1989; Ziatas, Durkin & Pratt 1998). Finally, it has been shown that even in young children with SLI, no ToM delay is detectible when the linguistic complexity of the task is kept low (Miller 2001), unlike what has been shown for ASD where ToM difficulties have thus been argued to be more central to their condition (Colle, Baron-Cohen & Hill 2007).

²⁰ As an illustration of legitimate clitic omission, consider the example from Fonagy (1985): While an ACC would be required in written French, a null object is acceptable because of the presence of a sufficiently salient discourse topic:

—Voulez-vous que je vous donne mon numéro de téléphone? —Non, je connais __.

—‘Want-you that I you give my number of telephone?’ — ‘No, I know.’

²¹ Given the highlighted relationship between perspective switching and performance with ACC1, we may speculate that other pronominals involving a shift in perspective could also be vulnerable in ASD, e.g., reflexive clitics elicited with a similar protocol to ours. As children with SLI perform well with these items (see, e.g., Tsimpli 2001; Tuller et al. 2011; Novogrodsky and Friedmann 2010), difficulty in ASD would once again indicate a difference in the language profiles of these populations. We leave this for future work.

²² This appears to be coherent with recent findings that distinct ASD subgroups (identified based on functional neuroimaging phenotypes in the first years of life) relate to different language trajectories, with those individuals with better language responding to speech by recruiting not only the canonical language areas of the brain but also the subcortical regions involved in emotion and memory, in a fashion more close to that observed in typically developing controls (Lombardo et al. 2015).