Complex syntax and working memory in children with specific learning difficulties

(1)

Article

Reference

Complex syntax and working memory in children with specific learning difficulties

STANFORD, Emily Nicole, DELAGE, Hélène

STANFORD, Emily Nicole, DELAGE, Hélène. Complex syntax and working memory in children with specific learning difficulties. First Language , 2019, Syntax and Verbal Short Term Memory in Developmental Disorders, p. 1-26

DOI : 10.1177/0142723719889240

Available at:

http://archive-ouverte.unige.ch/unige:149982

Disclaimer: layout of this document may differ from the published version.

1 / 1

(2)

https://doi.org/10.1177/0142723719889240 First Language 1 –26

sagepub.com/journals-permissions DOI: 10.1177/0142723719889240 journals.sagepub.com/home/fla

FIRST LANGUAGE

Complex syntax and working memory in children with specific learning difficulties

Emily Stanford and Hélène Delage

University of Geneva, Switzerland

Abstract

Working memory (WM) limitations are frequently reported for children with specific learning disorder (SLD). However, WM capacity influences more than literacy and numeracy, as research highlights the contribution of WM to language development, in particular syntax. In this article, the authors study the effect of syntactic intervention, i.e. syntactic elements intervening between filler and gap, on comprehension in children with SLD and evaluate the relationship of this effect to WM capacity. Specifically, they assess how these children comprehend wh-questions and relative clauses. Additionally, they examine how comprehension relates to WM, measured by backward digit recall.

The authors report that a subgroup of children with SLD struggle to comprehend structures containing intervention, and that WM capacity influences performance in SLD.

Their conclusion is that computing a syntactic relation in which a moved object and an intervening subject share a featural specification taxes the processing system of children with SLD who have WM limitations. Thus, syntactic difficulties, although not typically associated with SLD, may form part of the SLD profile, especially when WM capacity is reduced.

Keywords

Children, comprehension, intervention, SLD, working memory

Introduction

Specific learning disorder (SLD)

Specific learning disorder (SLD), one of the most frequently diagnosed developmental disorders, is characterized by unexpectedly poor attainment in key academic domains,

Corresponding author:

Emily Stanford, Department of Linguistics, University of Geneva, 5 Rue De-Candolle, 1205 Geneva, Switzerland.

Email: emily.stanford@unige.ch

Special Issue: Syntax and Verbal Short Term Memory in Developmental Disorders

(3)

such as literacy and numeracy, with prevalence rates of 3–10% for the former (Snowling, 2008) and 5–8% for the latter (Geary, 2004). The fifth edition of the Diagnostic Statistical Manual (DSM-5) (American Psychiatric Association, 2013) defines SLD as a neurode- velopmental disorder that manifests itself during the formal years of schooling as persistent difficulties in at least one of three primary academic domains: reading, writing and/

or mathematics. The specific learning difficulties are unforeseen as other areas of development appear to be unaffected, and they are often identified due to a discrepancy between the child’s actual achievement and expected achievement based on IQ measurements. Reid (2016) conceptualizes SLD as unexpected underachievement with a genetic etiology that reflects a set of core cognitive deficits across academic domains: attention difficulties, memory problems and poor processing speed. These pockets of weakness lead to a variety of symptoms that can range from mild to severe, such as inaccurate or slow and effortful reading, poor reading comprehension, written expression that lacks precision and clarity, difficulty remembering number facts, and inaccurate mathematical reasoning. Furthermore, learning difficulties often appear in comorbidity with other developmental disorders, such as attention difficulties (Margari et al., 2013), with recent research suggesting that the interaction of neurological processing mechanisms makes the co-occurrence of two disorders more probable than the occurrence of either disorder in isolation (Reid, 2016; Snowling, 2008).

While SLD is an all-encompassing label that allows clinicians to make a single diagnosis based on the unification of a range of typically co-occurring symptoms, children with SLD are often separated into more dichotomous categories that describe impairment in terms of one learning domain, such as literacy or numeracy, and which are considered subtypes of SLD. Developmental dyslexia (henceforth dyslexia) is the diagnostic term given to children who present enduring difficulties in literacy acquisition despite normal intelligence and adequate instruction, whereas developmental dyscalculia (henceforth dyscalculia) refers to difficulties in numeracy that impair one’s ability to acquire numerical competences and mathematical skills. In dyslexia, the dominant causal stance is that the disorder stems from a phonological deficit that affects how phonological rep- resentations are stored, retrieved and manipulated (Castles & Coltheart, 2004; Fawcett &

Nicolson, 2008; Ramus et al., 2003, 2013; Reid, 2016; Snowling, 2008; J. Stein, 2008), and dyscalculia is believed to result from a core deficit in numerical processing, which has been linked to the ability to retain, retrieve and manipulate verbal content over brief periods (Butterworth, 2010). While the main deficits found in the two disorders may seem unrelated, evidence suggests that similar operative cognitive competences are com- promised in both cases. More precisely, deficits in verbal working memory (WM) seem to give rise to both dyslexia and dyscalculia, an assumption that is supported by a recent meta-analysis (Peng & Fuchs, 2016). Indeed, there is strong evidence that WM influences the acquisition of reading and arithmetic as the encoding, storage and retrieval of information underpins many formal learning activities and is a reliable predictor of success in core academic areas (Gathercole & Alloway, 2008; Gathercole & Baddeley, 1990;

Gathercole & Pickering, 2000). It is thus unsurprising that children with poor WM profiles struggle to attain age-appropriate literacy and numeracy skills, as inability to meet the WM demands of learning situations results in missed learning opportunities and eventually learning delay. Limitations in WM are now consistently reported to be one of

(4)

the main cognitive deficits of children who significantly underperform at school (Gathercole & Alloway, 2008).

However, there is reason to believe that poor WM may influence more than the acquisition of literacy and numeracy in children with SLD, as robust empirical evidence highlights the important contribution of WM to various domains of language development, such as syntax (Adams & Gathercole, 2000; Delage & Frauenfelder, 2019; Engel de Abreu et al., 2011; Leclercq & Majerus, 2010; Majerus et al., 2009; Montgomery et al., 2008; Willis & Gathercole, 2001). In this study we investigate whether children with SLD show weakness in the comprehension of complex syntactic structures involving syntactic intervention, and how this may be related to limitations in WM as measured by backward digit span.

Syntactic complexity: wh-questions and relative clauses

It is a well-documented finding that the comprehension of object questions and object relative clauses is problematic for young children and several proposals have been put forward in an attempt to pinpoint what it is about these structures that makes them complex. Computational approaches, such as the Derivational Complexity Hypothesis (DCH) (Jakubowicz, 2011), describe complexity in terms of the number and nature of syntactic operations to be performed in order to arrive at the derived structure and link the mastery of computationally complex structures to cognitive constraints. More specifically, according to the DCH framework, the syntactic operation of displacing a syntactic element from its base position to one that is higher in the hierarchy (movement), is computationally complex and is bound by WM capacity. That movement operations are difficult for children has been empirically illustrated by acquisition studies of object questions in French, whose production data show that initially, children preferentially produce in situ (1) rather than ex situ structures (2) (Hamann, 2006; Jakubowicz, 2011;

Jakubowicz & Strik, 2008; van Kampen, 1997; Zuckerman & Hulk, 2001). In other words, children start by producing wh-questions that avoid movement, but with age adopt the more computationally complex structure that contains wh-fronting. The DCH interprets these findings as evidence that complex syntax emerges as WM capacity grows with maturation.

(1) Tu as poussé qui?

2p have pushed who

(2) Qui tu as poussé ?

who 2p have pushed

While movement operations increase complexity, movement alone does not render a structure complex as computationally comparable structures have been shown to gen- erate different processing patterns. For example, object-extracted which questions (3) are associated with poorer comprehension than object-extracted who questions (4) (Avrutin, 2000; De Vincenzi, 1991; Donkers & Stowe, 2006; Friedmann &

Novogrodsky, 2011; Sheppard et al., 2015) despite the fact that both structures are

(5)

derived via movement that fronts a wh-element. Grammatical approaches explain this asymmetry in terms of an intervention effect that stems from theories of locality, namely featural Relativized Minimality (f-RM) (Rizzi, 1990, 2001; Starke, 2001).¹ According to f-RM, a structure is difficult to compute when a lexically-headed object crosses over an intervening lexical subject, as the two elements share a +noun phrase (+NP) feature. The inclusion of the +NP feature in the intervening (c-commanding) lexical subject creates a filler–gap dependency that is costly to parse for the computational system as the relation, which is sensitive to distance, must be established across the embedded subject despite its featural similarity to the target head. In the case of object-extracted who questions, the featural specifications of the fronted wh-operator and the embedded lexical subject over which it crosses are distinct, thus the syntactic analysis is more easily realized.²

(3) Which mailman did the fireman pull _____?

[+Wh +NP] [+NP]

(4) Who did the fireman pull _____?

[+Wh] [+NP]

Evidence for an intervention effect as described by f-RM has been found across different age groups, different cognitive profiles and different languages (see Friedmann et al., 2009 for young typically developing [TD] children; Friedmann & Novogrodsky, 2011 for children with developmental language disorder [DLD]; and Garraffa & Grillo, 2008 and Sheppard et al., 2015 for individuals with aphasia). One thing these populations all have in common is known limitations in WM (see Gathercole et al., 2004 for TD children; Archibald & Gathercole, 2006 for children with DLD; and Majerus et al., 2015 for aphasia). Comprehension difficulties related to lexically-headed object structures may be due to the need to move a determiner phrase [DP] specified with a +NP feature over another DP with the same featural specifications, and to temporarily store both DPs in WM while computing the filler–gap dependency. In this case, the embedded lexical subject would behave as a memory distractor that competes with the displaced element for retrieval at the position of the gap as they share a similar retrieval cue. This is in line with models of cue-based retrieval parsing, in which memory retrieval of the filler at the position of the gap involves comparing a set of retrieval cues against other items in memory in parallel, thus an item that partially matches the retrieval cues of the target may be activated and erroneously retrieved. Formal linguistic principles such as f-RM make explicit what can be considered a cue in parsing, namely features, such as the +NP feature, that trigger movement. Differentiating the two DPs conceivably taxes WM, as computing a dependency in the presence of intervention would consume more processing resources than computing a dependency in which intervention is absent. For the still- developing or atypically-developing child system that lacks the necessary WM capacity to perform the syntactic computation, the processing mechanism would break down, impeding comprehension. It thus seems legitimate to hypothesize that syntactic computation is directly linked to processing, which is assumed to be at least partly sustained by WM, as formulated by Jakubowicz (2011).

(6)

Syntactic intervention and WM in comprehension

A number of studies have found an effect of WM resources on the accuracy of comprehension and on online measures of processing in TD children (Arosio et al., 2009, 2011, 2012; Booth et al., 2000) and in children with dyslexia (Robertson & Joanisse, 2010).

According to Baddeley and Hitch’s (1974) WM model, WM refers to the cognitive ability to temporarily hold and transform information during an ongoing mental task and includes three major components: (1) the phonological loop, which stores verbal information, (2) the visuospatial sketchpad, which stores visuospatial information, and (3) the central executive, a limited-resource attentional control system that coordinates the two slave systems. One advantage of Baddeley’s multicomponent model is that simple cognitive activities, such as storing verbal items in the phonological loop, can be dis- sociated from more complex cognitive activities that involve the supervisory central executive system, such as storing verbal items while performing an additional processing task. Barrouillet et al. (2007) suggest that the most accurate way to assess phonological loop capacity is through simple span tasks, such as the repetition of nonwords or the forward repetition of digits, and that central executive capacity is best measured through complex span tasks, such as the backward repetition of digits or listening span tasks.

With regard to the relationship between WM and the comprehension of structures containing syntactic intervention, empirical evidence suggests that both simple and complex span play a role, in both typical and atypical development. In what concerns typical development, Booth et al. (2000) showed that forward digit span impacted on online measures of relative clause processing in English-speaking children aged 8 to 11: children with high forward digit span capacity were associated with higher accuracy rates on self-paced reading and listening tasks. It should be noted, however, that in the same study, a significant link was not found between comprehension and complex span, measured by a reading span task. Also using a self-paced reading paradigm, similar results were obtained by Arosio and colleagues (2011), who reported that forward digit span significantly affected the comprehension accuracy of object relative clauses in school- aged Italian-speaking children (mean age: 9;33), and that this effect was modulated by animacy. By contrast, these authors found no significant effect of complex span, as measured by a listening span task, on the comprehension of object relative clauses. However, when a picture selection task was used to test the comprehension of object relative clauses and its relation to backward digit span (complex span) in Italian-speaking children aged 5 to 11, a significant correlation was reported (Arosio et al., 2009). In German- speaking 7-year-olds, Arosio and colleagues (2012) also used a picture selection task to show that forward digit span (simple span) had a significant effect on the comprehension of object relatives: children with higher span capacity were associated with better performance on the comprehension task. Finally, in one of the few studies that has investigated the relationship between the comprehension of complex syntax and WM in children with learning difficulties, Robertson and Joanisse (2010) reported that in 14 children aged 9 to 12 who had been diagnosed with dyslexia, the processing of object relative clauses demonstrated by a picture selection task was significantly influenced by simple span capacity, measured by forward digit recall.

(7)

To summarize, there is strong evidence that WM capacity, measured by simple and complex span, is fundamentally related to the comprehension of complex syntax. While this does not necessarily establish a cause–effect relation, it seems relevant to investigate whether children with SLD, who have known WM deficits, present significant difficulties in syntax. Studies focusing specifically on dyslexia indicate that this is indeed the case, with Scarborough (1990) reporting that children who were at risk of dyslexia pro- duced shorter and less complex utterances at 30–40 months. This suggests that the learning of letter–sound correspondences in school-aged children and the processing of morphosyntax utilize similar underlying cognitive mechanisms, namely the verbal component of WM. Therefore, initial syntactic skills may presage literacy difficulties.

Additionally, crosslinguistic evidence from older children shows that individuals with dyslexia perform poorly on tough constructions, such as relatives and passives, when compared to age-matched controls (see Leikin & Bouskila, 2004 for Hebrew; Talli et al., 2013 for Greek; Bar-Shalom et al., 1993; Robertson & Joanisse, 2010; C. L. Stein et al., 1984 for English; Arosio et al., 2017 and Cardinaletti & Volpato, 2015 for Italian; and Sevcenco et al., 2014 for Romanian). To our knowledge, no research has investigated the possibility that children with dyscalculia demonstrate difficulties in syntax, although it would be reasonable to assume this is the case if dyslexia and dyscalculia share a common deficit in verbal WM.

Aim of current study

Several studies have reported a subject–object comprehension asymmetry (Bar-Shalom et al., 1993; Mann et al., 1984; C. L. Stein et al., 1984), and the difficulties associated with object structures have been linked to the movement of an object across a subject intervener (e.g. Arosio et al., 2017). Robertson and Joanisse (2010) found that in children with dyslexia, simple span correlated with comprehension accuracy in high-demand storage and processing conditions, such as when they were asked to process object relative clauses (ORs). However, children with SLD have been associated with both simple and complex span limitations, and Helland and Asbjørnsen (2004) showed that children with literacy deficits and children with comorbid literacy and numeracy deficits were more likely to be impaired in complex span (measured by the backward recall of digits) than in simple span (measured by the forward recall of digits). Furthermore, empirical evidence suggests that a specific relationship exists between complex span and the comprehension of complex syntax, and that complex span capacity might be a better indica- tor of how complex structures are processed in atypical populations, such as in children with DLD (Delage & Frauenfelder, in press; Frizelle & Fletcher, 2015; Montgomery &

Evans, 2009; Montgomery et al., 2008). For this reason, we focus exclusively on complex span in this study.

While complex span deficits have been systematically identified in children with SLD, as far as we know, no studies have investigated the relationship between performing complex syntactic operations (e.g. creating a long-distance dependency across an intervener) and complex span in this population. Although children with SLD form a heterogeneous group, we justify combining the subtypes of SLD as we accept the prem- ise that both dyslexia and dyscalculia stem from the same set of core cognitive deficits,

(8)

in particular deficits in verbal WM. Motivated by principles and properties established in theoretical linguistic literature, cue-based models of comprehension and empirical findings related to complex span capacity and syntactic performance, the current study aims to examine whether children who underperform scholastically show weakness in the comprehension of complex syntactic structures containing a syntactic intervener and how this may be related to weakness in WM, particularly in complex span. We hypothesize that children with SLD will find it difficult to compute syntactic intervention, impeding comprehension, and that poor comprehension will correlate with complex span deficits.. Such findings would be theoretically significant as they would further elucidate what is already known about WM and grammar vulnerability in children with dyslexia, and they would shed new light on this relationship in dyscalculia. These findings would also be clinically relevant as they would highlight the importance of syntactic remedia- tion in children with SLD.

Two referent decision tasks were used to study the effect of intervention on comprehension in school-aged Anglophone children with documented learning difficulties, and to evaluate the relationship of this effect to complex span. Concerning syntax, based on what has been observed in studies on TD children (Friedmann et al., 2009) and children with DLD (Friedmann & Novogrodsky, 2011), we predict that (1) a significant intervention effect will be found in the accuracy measurements of children with SLD, and that the computational complexity associated with the presence of intervention will lead to a decrease in accuracy and (2) the intervention effect will be stronger for children with SLD than it is for TD children, as limited WM resources in the SLD group should influence how these children process complex syntax. Concerning WM, we predict that (1), as was found by Helland and Asbjørnsen (2004), children with SLD will have complex span limitations, demonstrated by poor backward digit span performance, and (2) complex span and the comprehension of structures containing NP-intervention will be correlated in SLD, as was the case for DLD (Delage & Frauenfelder, in press; Montgomery

& Evans, 2009; Montgomery et al., 2008).

Method Participants

Eleven native English-speaking children (10 monolingual, 1 bilingual) who had been formally diagnosed with SLD and whose ages ranged from 7;8 to 11;6 (M = 9;8, SD = 1;1) participated in this study. All SLD participants received regular learning support and were primarily recruited through specialist schools in Switzerland (n = 5) and in the USA (n = 6). Inclusion criteria were an SLD diagnosis by a qualified educational psy- chologist and mother tongue mastery of English (exposure to English from birth). Of the 11 children, the majority had been diagnosed with dyslexia, more than half of the children had comorbid attention deficit hyperactivity disorder (ADHD), and one of the participants had comorbid coordination difficulties (dyspraxia). One participant had been diagnosed with dyscalculia, and one other participant had been diagnosed with SLD with a specific deficit in reading comprehension. Figure 1 outlines the distribution of learning difficulties in this group.

(9)

To provide age-matched controls for the children with SLD, 11 native English- speaking TD children aged 7;4 to 11;2 (M = 9;4, SD = 1;6) were also tested, either in Switzerland (n = 6) or in the USA (n = 5). The SLD and TD age means did not differ (p = .6). Of the children in the TD group, seven were monolingual, three were bilingual and one was trilingual. All TD children attended conventional schools and, according to their parents and teachers, had no history of learning difficulties and had never needed the targeted help of specialized learning support. Participant group data are presented in Table 1.

For all participants recruited in Switzerland, both SLD and TD, English was the main or only language spoken at home. Language of education for these participants, however, Figure 1. Distribution of learning difficulties in the SLD group.

Table 1. Overview of participant information: mean age and SDs in years and months, gender of participants, and number of languages spoken by participants.

Group Number of

participants Age Gender Lingualism

Mean (y;m) SD F M Mo. Bi. Tri.

SLD 11 9;8 1;1 2 9^a 10 1 0

TD 11 9;4 1;6 7 4 8 2 1

aThe majority of the participants in the SLD group were boys, highlighting the higher prevalence of SLD among males than females. Researchers, however, disagree on whether or not this is due to actual biologi- cal vulnerability, or simply referral bias (Hallahan & Kauffman, 2003).

(10)

tended to differ based on cognitive group: almost all children with SLD tested in Switzerland were receiving an Anglophone education and almost all TD children tested in Switzerland were receiving a Francophone education. This is likely due to the fact that the children with SLD were recruited from a learning support school opened especially for native English-speaking children with learning difficulties in the Geneva area. Parents of TD participants in Switzerland tended to favor a Francophone education, and many of these children were sequential bilinguals whose first exposure to French occurred at the start of their primary education. All participants recruited in the USA were monolingual English speakers and received a monolingual English education. The Appendix summarizes the participants’ language backgrounds.

This study was approved by the ethics committee at the University of Geneva and all parents signed a consent form before the study began.

Material and procedure

Complex span assessment. To measure the children’s ability to temporarily store and manipulate verbal information, complex span was assessed in both SLD and TD participants using the backward digit component of the digit memory test, a WM subtest of the Wechsler Intelligence Scale for Children Fourth Edition (WISC-IV) (Wechsler, 2003).

For this task, each participant was verbally presented with an item containing a series of digits (e.g. 2, 7) that were to be immediately recalled in the reverse order as read aloud by the examiner. Item length grew successively and the testing ended when the participant failed to correctly repeat two consecutive items of the same length. Each child received a backward digit score that was matched to a standard score, and which placed the child on a ranked scale for complex span capacity. For this study, scores equal to or below the 10th percentile were considered to represent impairment. This decision was based on the discrepancy criteria recommended in the DSM-5 (American Psychiatric Association, 2013).

Comprehension tasks I and II

Animations. To evaluate the comprehension of subject- and object-extracted struc- tures, we devised two different comprehension tasks: one to test wh-questions and one to test relative clauses. Eighteen animations were created using Anime Studio Debut 10 from Smith Micro Software. For these animations, three categories of figures, female, male and animal, were designed, and each category contained three types (e.g. male:

policeman, fireman and mailman). Each animation depicted three aligned figures in which one of the peripheral figures performed an action corresponding to an agentive transitive verb (e.g. pull) on the middle figure and the middle figure performed the same action on the other peripheral figure. The direction of the action was randomized but counterbalanced so that it proceeded from left to right in half of the animations and from right to left in the other half. For the actions that proceeded from right to left, the figure on the right always represented the agent and the figure on the left always represented the theme. The opposite was true for actions occurring in the reverse direction. Figure 2 shows the timeline of one of the right-to-left animations. Only figures from the same category appeared together in the animations, and the two peripheral figures were always of

(11)

the same type (e.g. policeman – fireman – policeman) but differed in one feature, such as color of clothing. Each figure appeared in each possible position (left – central – right) throughout the experiment.

Tasks I and II. The comprehension of subject- and object-extracted structures was assessed using two different tasks: task 1, the wh task, required the participants to respond to either subject or object who and which questions (6 to 9), and task 2, the relative clause (RC) task, required the participants to respond to commands containing either a subject relative or an object relative (10 and 11 respectively).

(6) Who pulled the fireman?

(7) Who did the fireman pull?

(8) Which policeman pulled the fireman?

(9) Which policeman did the fireman pull?

(10) Point to the policeman that pulled the fireman.

(11) Point to the policeman that the fireman pulled.

For both comprehension tasks, the participants watched an animation (refer to Figure 2 for the timeline of one of the animations), heard either an audio recorded wh-question or an audio recorded command while looking at a blank screen, and responded while looking at a static image of the three figures in their original positions. The participants responded by pressing one of two designated keys, one corresponding to the agent of the action and the other corresponding to the theme, which varied according to the direction of the animation. In the wh task, every animation was repeated four times so that the participants were presented with a total of 72 randomized animations: 18 followed by Figure 2. Timeline of a right-to-left animation as seen in both comprehension tasks.

(12)

who subject questions, 18 followed by who object questions, 18 followed by which sub- ject questions and 18 followed by which object questions. The which object condition was the only one to contain intervention in this task. In the RC task, each animation was repeated twice so that the participants were presented with a total of 36 randomized animations: 18 followed by a command containing a subject relative, the non-intervention condition, and 18 followed by a command containing an object relative, the intervention condition. All test items were experimental as neither task contained fillers. Before each task, the children took part in a brief training session to ensure that they could recognize the characters and to get familiarized with the procedure. Including training, the entire experiment took 90 minutes and was either completed in one session with several pauses, or in two sessions of 45 minutes. Table 2 gives examples of test items based on the animation seen in Figure 2.

Results

As our experimental design asked participants to press one of two designated keys, we wanted to exclude any children who displayed random performance by performing at chance level in both the subject and the object condition. Binomial analysis determined that none of the children in either group (SLD/TD) displayed random performance. Due to non-conformity to normality of distributions (ascertained by a Shapiro–Wilk test), non-parametric statistics are reported throughout this article. We used the Mann–Whitney test for inter-group comparisons between the SLD and TD participants and the Wilcoxon test for intra-group comparisons within the SLD group. The Spearman rank-order correlation coefficient was used to measure the strength of the relationship between complex span and syntax.

Syntax

An intervention effect has been found in young TD children and in school-aged children with DLD, thus we expected to find a significant intervention effect, characterized by a drop in comprehension accuracy, in school-aged children with SLD. To determine if this effect was indeed present in SLD, intra-group comparisons were made by analyzing how accuracy varied as a function of intervention in the two comprehension tasks. For the wh task, the results in Figure 3 indicate a significant difference between wh-questions in which intervention was absent (−INT) and those in which intervention was present (+INT): accuracy was at ceiling for −INT wh-questions (both types of subject questions and who object questions), but dropped significantly for +INT wh-questions (which object questions). Likewise, for the relative clause task, subject relatives were associated with a high level of accuracy, while object relatives were associated with a significantly poorer performance. These results are summarized in Table 3.

Next, to determine whether children with SLD demonstrate impaired comprehension when intervention is present, inter-group analyses compared SLD performance to TD performance for the intervention conditions (which object questions in the wh task and ORs in the RC task). For the wh task, the results showed no significant SLD–TD difference in accuracy measurements, U = 33, p = .076, although there was a clear tendency

(13)

Table 2. Examples of test items for tasks 1 and 2.

Wh task Subject wh-questions Object wh-questions Audio recording:

Which policeman/who pulled the fireman?

Expected response:

Audio recording:

Which policeman/who did the fireman pull?

Expected response:

RC task Subject relatives Object relatives

Audio recording:

Point to the policeman that pulled the fireman.

Expected response:

Audio recording:

Point to the policeman that the fireman pulled.

Expected response:

(14)

for the SLD group to be less accurate than the TD participants. For the RC task, a group effect was revealed as the SLD–TD difference in the +INT condition was significant, U

= 19, p = .005, with the TD group outperforming the SLD group. When intervention was absent, SLD and TD children behaved similarly, with both groups of children performing at ceiling on both tasks (wh task = 92% accuracy for children with SLD and 99% accuracy for TD children; RC task = 92% accuracy for children with SLD and 98%

accuracy for TD children). These findings are summarized in Table 4.

While the results reveal that children with SLD tend to demonstrate poorer comprehension of +INT structures than age-matched control children, it is important to take precautions with group performance because upon closer inspection there was a high degree of variability within this group: some children with SLD performed very well and others performed poorly. To investigate if a subgroup of children with SLD has problems Figure 3. Response accuracy in comprehension of wh-questions in SLD children.

Table 3. Comparison of the comprehension accuracy percentages of +/−INT wh-questions and relative clauses in children with SLD.

−INT +INT −INT vs. +INT

Wh task 92.3%

(SD = 7.8%) 73.7%

(SD = 32.9%) p = .022

RC task 92.4%

(SD = 15.2%) 80.8%

(SD = 23.7%) p = .028

wh task, −INT = who subject/object questions, which subject questions.

wh task, +INT = which object questions.

RC task, −INT = subject relative clauses.

RC task, +INT = object relative clauses.

(15)

with +INT structures, we performed an individual analysis by examining the proportion of children with SLD that had difficulties comprehending structures containing intervention. Replicating analyses that have been performed in studies on relative clauses in dyslexia (Guasti et al., 2015), we examined individual SLD accuracy measurements in +/−INT questions and relative clauses, and considered children who scored 1.5 SDs or more below the TD mean but were still above chance level as having subtle comprehension difficulties.³ Scores that were at or below chance level were taken to indicate more severe comprehension difficulties. For the wh task, five of the 11 children with SLD were at least 1.5 SDs below the mean of the TD group on which object questions, and for the RC task, eight of the children with SLD were at least 1.5 SDs below the TD mean for object relatives. Furthermore, two children with SLD performed below chance level in the +INT condition for both tasks, with one of them also performing at chance level in the subject condition of the RC task. Table 5 summarizes these findings.

Working memory

Concerning WM, children with SLD were expected to have complex span limitations, demonstrated by poor performance on the backward digit memory span task when compared to their TD age-matched peers. In the SLD group, two children performed within the impaired range (< 10th percentile), with a third child ranking in the 11th percentile.

By contrast, all TD children had complex span measurements that fell within the normal range (⩾ 10th percentile) (see Figure 4 for the distribution of percentile rankings for backward digit span for the two groups). The results of the Mann–Whitney test reported no statistically significant difference between the two groups for backward digit span, U = 34, p = .088, although there was an important tendency for percentile rankings to be lower in the SLD group (M = 51, SD = 34) than in the TD group (M = 76, SD = 21).

Considering there was a high degree of variability in the SLD performance on the WM task, as there was for syntax, we investigated if a particular subgroup of children with SLD has complex span limitations. This was done by considering each individual SLD score and examining how it deviated from the TD mean. Five of the 11 SLD participants scored at least 1.5 SDs below the TD mean (summarized in Table 5), and the children who performed at the low end of the spectrum on the digit memory test tended to perform at the low end of the spectrum on the comprehension tasks. Of the six children with SLD who performed at the mid to high end of the spectrum on the digit memory Table 4. Comparison of SLD/TD accuracy percentages when intervention is present (which object questions and object relative clauses).

SLD [+INT] TD [+INT] SLD vs. TD

Wh task 73.7%

(SD = 32.9%) 92.9%

(SD = 9.7%) Tendency

RC task 80.8%

(SD = 23.7%) 99.5%

(SD = 1.7%) p = .005

(16)

Table 5. Individual SLD scores in the two comprehension tasks and percentile rankings for backward digit span. In bold are scores that are 1.5 SDs or more below the TD mean, while scores highlighted in grey represent performance that is at or below chance level.

SLD participants Diagnosis Wh task RC task Complex

span (percentile ranking) ACC %

[+INT] ACC %

[−INT] ACC %

[+INT] ACC % [−INT]

RT Dyslexia,

dyspraxia, ADHD 78 89 83 83 67

BR Dyslexia 100 100 100 100 97

JF Dyslexia 89 100 100 100 96

LV Dyslexia 100 96 83 100 57

MV Dyslexia 72 94 94 100 8

CH RCD, ADHD 22 80 33 50 11

MH Dyslexia, ADHD 94 98 100 100 50

BK Dyslexia, ADHD 94 96 94 100 68

LD Dyscalculia 89 87 89 89 21

CS Dyslexia, ADHD 0 78 39 94 5

PK Dyslexia, ADHD 72 96 72 100 24

TD participants

Mean 93 99 100 98 76

ACC = accuracy.

Figure 4. Distribution of percentile rankings for backward digit span for the SLD and TD participants.

(17)

test, five also had high accuracy scores on the comprehension tasks. One child, however, performed well on the digit memory test but was significantly below the TD mean for performance on the comprehension tasks.

Finally, an important aim of our work was to explore the links between WM, especially complex span, and the comprehension of structures containing intervention in children with SLD. In order to do that, correlation analyses and simple linear regression were performed. For the intervention condition, the results showed a positive, statistically significant correlation between accuracy measurements and complex span for both the wh task, r = .73, p = .011, and the RC task, r = .63, p = .037. Simple linear regression corroborated the correlation results: higher complex span scores predicted higher accuracy scores for the wh task, β = .75, t(9) = 3.37, p = .008. Complex span also explained a significant proportion of variance in the accuracy scores, R² = .51, F(1,9) = 11.37, p = .008. The same was true for the RC task, β = .70, t(9) = 2.96, p = .02; R² = .44, F(1,9) = 8.75, p = .02.

Concerning the non-intervention condition, a positive, statistically significant correlation was only found between accuracy and complex span in the wh task, r = .62, p = .042. Simple linear regression demonstrated a tendency for complex span to predict accuracy scores, although this was not a statistically significant model, β = .59, t(9) = 2.17, p = .058; R² = .27; F(1,9) = 4.72, p = .058.

Discussion

This study made four main predictions: (1) that in children with SLD comprehension would be poorer for +INT structures than for −INT structures, (2) that comprehension accuracy of +INT structures would be lower for children with SLD than for TD children, (3) that children with SLD would demonstrate weakness in WM as measured by backward digit span, and (4) that in SLD, the degree of sensitivity to intervention would be linked to digit span (i.e. poor comprehension would be associated with low span capacity). To answer these questions, we assessed the comprehension of lexically-headed object structures in school-aged TD children and children with SLD because research has shown that structures containing NP-intervention are computationally complex and an excellent tool for detecting grammar difficulties. Additionally, based on empirical evidence that there is a link between complex syntax and WM, in particular complex span in atypical populations, we examined the relationship between backward digit span and the comprehension of wh-questions and relative clauses in SLD.

Syntax in SLD: The effect of intervention

Our results confirm that in 7- to 11-year-old children with SLD, comprehension is affected by syntactic complexity, characterized in terms of intervention and featural similarity: these children easily comprehended −INT structures (subject relatives, who/

which subject questions and who object questions) whereas +INT structures (lexically- headed object relatives and which object questions) were problematic and were associated with a drop in accuracy. These findings corroborate what has been observed

(18)

crosslinguistically and across various age groups and cognitive profiles, and offer additional support for the hypothesis that feature similarity plays a significant role in sentence comprehension. The fact that who object questions are unproblematic for children with SLD confirms that moving an object across a subject is not the only source of complexity in comprehension, but that comprehension difficulties emerge when the moved object and the intervening subject both contain the same grammatically-active feature, in this case +NP.

Another finding of the present study was that children with SLD are not only sensitive to intervention, but this sensitivity is outside of the typical range as children with SLD performed poorly when intervention was present, as opposed to their TD peers whose performance remained at ceiling regardless of the presence or absence of intervention.

The observed ceiling effect for TD children most likely signifies that the tasks were too easy for school-aged TD children (indeed, previous studies on intervention in typical development have tested much younger cohorts), whereas there was still a clear asymmetry in the comprehension of +/−INT structures in the SLD group. Furthermore, exam- ination of individual performance for accuracy suggests that a subgroup of children with SLD has greater difficulty than TD children in successfully performing complex syntactic operations involving intervention, although this difficulty seems to range from subtle to quite severe. In the case of severe comprehension difficulty, performance resembles what has been reported for children with DLD (Contemori & Garraffa, 2010; Håkansson

& Hansson, 2000; Novogrodsky & Friedmann, 2006). These findings are not unexpected as nine of our 11 SLD participants were diagnosed with dyslexia, and the dyslexia–DLD overlap is well documented (Bishop & Snowling, 2004; Catts et al., 2002, 2005; Flax et al., 2003; McArthur et al., 2000). Indeed, when McArthur and colleagues evaluated the oral and written language capacity of 110 English-speaking children aged 7 to 13 who had been diagnosed with dyslexia, they found that 55% of these children also presented an impairment in oral language and could be classified as having DLD. Similarly, when 102 children with DLD were tested in the same study, 51% of them displayed deficits in written language. In other words, over half of the children in McArthur and colleagues’ study met the criteria for both disorders. Why is it the case that the syntactic difficulties of so many children diagnosed with literacy impairment go undiagnosed? It is possible that failure to evaluate the oral language abilities of children presenting symptoms of dyslexia can lead to under-identification of DLD. If assessment is recommended because a child is having difficulty learning to read, write or spell, more subtle difficulties in syntax may be overlooked. It has also been suggested that some portion of children with DLD may in fact be misdiagnosed as having dyslexia, or that a subgroup of children with dyslexia may also be affected by unidentified comorbid DLD (Guasti et al., 2015). We interpret our results similarly: although the difficulties reported for our SLD population prior to our study were limited to the domains of literacy and numeracy, a number of our SLD participants demonstrated subtle deficits in syntax that may indicate the presence of unidentified DLD, whether misdiagnosed or co-occurring. However, as our study did not include a comparison group of children with DLD, this interpretation remains speculative as qualitative and quantitative similarities between SLD and DLD could not be investigated.

(19)

Relation between WM and syntax

Why do children with SLD struggle to comprehend structures containing intervention whereas such structures are unproblematic for their TD counterparts? As previously described, WM capacity is closely linked to syntactic performance, and children with SLD are known to have simple and complex span limitations. Jakubowicz (2011) hypothesizes that the persistent grammar difficulties observed in DLD, such as the poor comprehension of +INT structures, is the result of span weaknesses that limit one’s ability to process complex syntax, a claim that is supported by empirical evidence (Montgomery & Evans, 2009; Montgomery et al., 2008). Otherwise stated, immature WM development is assumed to be at least partly responsible for immature grammar in DLD. It is possible that a similar conclusion can be made for SLD. If span development does not follow a typical, age-dependent trajectory in SLD, this might account for the atypical comprehension patterns seen in this population when compared to age-matched TD peers.

We used the digit memory test to investigate span capacity in children with SLD.

Because complex span has been closely linked to the comprehension of complex syntax, especially in atypical populations (Delage & Frauenfelder, in press; Frizelle &

Fletcher, 2015; Montgomery & Evans, 2009; Montgomery et al., 2008), we chose to focus exclusively on backward digit span performance, predicting that the SLD participants would demonstrate impaired performance when compared to the TD group. This, however, was not the case as the two groups could not be statistically distinguished based on global complex span performance. This finding is unexpected as it contradicts what has been reported for children with literacy and numeracy difficulties (Helland &

Asbjørnsen, 2004). Our digit span assessment did, however, reveal different TD and SLD complex span profiles. SLD scores were widely distributed and could be classified as impaired to above average, with scores in this group ranging from the 5th to the 97th percentile. Using a 10th-percentile cut-point to differentiate impaired from pre- served complex span in children with SLD, eight children were within the normal range (⩾ 10th percentile) and only two demonstrated significant complex span impairment (< 10th percentile). However, when we considered SLD scores falling below 1.5 SDs of the TD mean, we identified two distinct subgroups: five children with SLD whose performance was significantly poorer than that of the TD children and six children with SLD who performed similarly to the TD group. TD scores showed a much narrower distribution, and all individual TD participants scored within 1.5 SDs of the TD mean.

These findings highlight the important inter-individual variability in complex span capacity within children with learning difficulties and indicate that while limitations in complex span are systematically associated with academic under-attainment, they may not be a condition sine qua non of SLD. Unsurprisingly, the five children with SLD who had the lowest complex span scores tended to be the children with the lowest comprehension scores, and the six children with SLD who had the highest complex span scores tended to be the children with the highest comprehension scores. The only exception was one child who performed well on the complex span task but was significantly below the TD average on the comprehension tasks. However, it should be noted that this child had comorbid attention difficulties and found it particularly difficult to maintain

(20)

focus during the comprehension tasks, which took place after the span task and took more time to complete than the span task. It is also important to keep in mind that our WM measurements were limited to a singular evaluation. While the digit memory test is widely used for measuring simple and complex memory span and backward digit span has been shown to predict syntactic capacity in children with DLD (Delage &

Frauenfelder, in press; Durrleman & Delage, 2016), pairing it with other complex span tasks, such as the running span task (Pross et al., 2008) or the counting span task (Danahy et al., 2007), would have perhaps given a more precise indication of complex span capacity in children with SLD.

What, then, differentiates SLD from TD children if they cannot be reliably distinguished on the basis of WM capacity alone? One possibility is that it is not WM capacity, i.e. the size of complex span, that plays a significant role in the comprehension of lexically-headed object structures, but rather weakness in some other memory mechanism that influences processing. For example, if we assume that successful memory retrieval (and consequently successful comprehension) depends on the number of items in memory that match retrieval cues, one could hypothesize that during dependency resolution in which an intervening lexical subject competes with a displaced lexical object for retrieval at the position of the gap, children with SLD implement memory retrieval cues differently than TD children. More specifically, children with SLD may have more difficulty inhibiting a distractor (i.e. an intervener) when monitoring incom- plete dependencies, and may thus show increased susceptibility to interference (i.e.

intervention) during memory retrieval operations in sentence processing. In other words, SLD–TD differences in the comprehension of +INT structures may be related to individual differences in success of memory retrieval operations (e.g. inhibiting memory distractors) and not individual differences in processing capacity (e.g. span size). If it is indeed the case that children with SLD struggle to inhibit interference when resolving filler–gap dependencies, this would support arguments put forward by Cunnings (2017) on sentence processing in sequential bilinguals, another population known to have difficulty computing complex syntax (Marinis & Saddy, 2013). This also opens up a number of new research questions about the role of executive functions other than WM in the comprehension of complex syntax. It is a well-documented finding that children with ADHD struggle with inhibitory control (Barkley, 1997; Scheres et al., 2004; Wodka et al., 2007), and over half of our participants with SLD had comorbid ADHD. Furthermore, children with attention difficulties demonstrated poor performance on our comprehension tasks, and children with both limited WM capacity and attentional difficulties were among those with the most impaired comprehension.

However, as our study did not include a measure of inhibition, it is not possible to judge the magnitude of the potential effect that it has on syntactic computation and processing. New studies should investigate how inhibition influences performance on syntactic tasks in children with SLD, especially those that involve the computation of a dependency in the presence of intervention.

Finally, to investigate the relationship between the comprehension of complex syntax and complex span capacity in SLD, we performed correlation analysis. Our results con- firmed our prediction that the two are closely related, with lower complex span capacity being associated with less accurate performance in both comprehension tasks. These

(21)

findings are in line with research that has reported a close relationship between complex span and complex syntax in children with DLD (Delage & Frauenfelder, in press) and suggest that, similarly to children with DLD, complex span weakness in children with SLD limits the ability to compute complex syntactic operations, in particular in structures containing an NP-intervener. Thus, not only do our results support syntactic theory that describes performance differences associated with certain syntactic structures in terms of a distinctness hierarchy of featural specifications that constrain syntactic com- putations, they also support a WM model of cue-based retrieval parsing. It is also worth noting that we found a correlation between performance on the complex span task and performance on the non-intervention condition of the wh task. This is not necessarily surprising as this condition contained who object questions, in which the canonical order of the arguments is changed, and who/which subject questions, which are also derived via movement operations that displace the head NP. We know that noncanonicity and movement operations increase complexity, need WM resources to be performed, and can be difficult to master in atypical development, even when intervention is absent. Indeed, Novogrodsky and Friedmann (2006) reported that 9- to 15-year-old Hebrew-speaking children with DLD fare considerably better with non-intervention structures than with intervention structures, but that non-intervention structures are still impaired when performance is compared to that of TD children. While our SLD participants were almost all at ceiling for the comprehension of non-intervention wh-questions, there was nonethe- less more variability for this measure than for the non-intervention relative clause measure. Furthermore, two participants with particularly low complex span scores also had low comprehension scores for non-intervention questions, which certainly influenced this correlation.

The findings that an intervention effect is present in English-speaking children with SLD and that this effect is stronger in children with WM limitations have clear clinical implications. If WM limitations are indeed responsible for deficits computing complex syntax in SLD, it may be of interest to train WM in this population to explore the possibility that WM rehabilitation positively influences syntactic performance. In fact, such a training study with promising preliminary results is currently underway with children with DLD (Delage et al., 2017; Stanford et al., 2019). Considering the similarities between children with DLD and a subgroup of children diagnosed with SLD, it would not be unreasonable to test the efficiency of such a training program in both populations.

Additionally, we think it is important for future work to examine the relationship between other executive functions, such as inhibition, and language acquisition in SLD in order to better understand the behavioral profile of this population.

Acknowledgments

The authors would like to thank the parents and children who participated in this project, as well as Oak Hill in Nyon, Switzerland, The Hill Center in Durham, NC and Chatham Academy in Savannah, GA.

Author contributions

ES wrote the manuscript, tested the participants and conducted all statistical analyses. Both authors reviewed the final manuscript.

(22)

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/

or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Emily Stanford https://orcid.org/0000-0002-6294-2410 Notes

1. We adopt the following definition of f-RM: In the configuration . . .X. . .Z. . .Y. . ., a local relation cannot hold between X and Y when

a. Z intervenes between X and Y, and

b. Z fully matches the relevant featural specification of X

2. It should be pointed out that not all features play a role in the computation of similarity.

Belletti et al. (2012) showed that only features that incite syntactic movement are relevant for intervention effects, and that a particular feature, such as gender, can have a different status crosslinguistically. The effect of featural mismatch on comprehension is outside of the scope of this study, but see Belletti et al. (2012) for a detailed explanation of this phenomenon.

3. It should be noted that the TD mean was rather high with very little variance in the data. This means that some children with SLD performed close to ceiling, while still scoring at least 1.5 SDs below the TD mean.

References

Adams, A. M., & Gathercole, S. E. (2000). Limitations in working memory: Implications for language development. International Journal of Language & Communication Disorders, 35(1), 95–116.

American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (5th ed.).

Archibald, L. M., & Gathercole, S. E. (2006). Short-term and working memory in specific language impairment. International Journal of Language & Communication Disorders, 41(6), 675–693.

Arosio, F., Adani, F., & Guasti, M. T. (2009). Grammatical features in the comprehension of Italian relative clauses by children. In J. M. Brucart, A. Gavarrò, & J. Sola (Eds.), Merging features:

Computation, interpretation and acquisition (pp. 138–155). Oxford University Press.

Arosio, F., Guasti, M. T., & Stucchi, N. (2011). Disambiguating information and memory resources in children’s processing of Italian relative clauses. Journal of Psycholinguistic Research, 40(2), 137–154.

Arosio, F., Panzeri, F., Molteni, B., Magazù, S., & Guasti, M. T. (2017). The comprehension of Italian relative clauses in poor readers and in children with specific language impairment.

Glossa, 2(1), 1–25.

Arosio, F., Yatsushiro, K., Forgiarini, M., & Guasti, M. T. (2012). Morphological information and memory resources in children’s processing of relative clauses in German. Language Learning and Development, 8(4), 340–364.

(23)

Avrutin, S. (2000). Comprehension of discourse-linked and non-discourse-linked questions by children and Broca’s aphasics. In Y. Grodzinsky, L. P. Shapiro, & D. A. Swinney (Eds.), Language and the brain (pp. 295–313). Academic Press.

Baddeley, A., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), Psychology of learning and motivation (Vol. 8, pp. 47–89). Academic Press.

Barkley, R. A. (1997). Behavioral inhibition, sustained attention, and executive functions:

Constructing a unifying theory of ADHD. Psychological Bulletin, 121(1), 65–94.

Barrouillet, P., Bernardin, S., Portrat, S., Vergauwe, E., & Camos, V. (2007). Time and cognitive load in working memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33(3), 570–585.

Bar-Shalom, E., Crain, S., & Shankweiler, D. (1993). A comparison of comprehension and production abilities of good and poor readers. Applied Psycholinguistics, 14(2), 197–227.

Belletti, A., Friedmann, N., Brunato, D., & Rizzi, L. (2012). Does gender make a difference?

Comparing the effect of gender on children’s comprehension of relative clauses in Hebrew and Italian. Lingua, 122(10), 1053–1069.

Bishop, D. V., & Snowling, M. J. (2004). Developmental dyslexia and specific language impairment: Same or different? Psychological Bulletin, 130(6), 858–886.

Booth, J. R., MacWhinney, B., & Harasaki, Y. (2000). Developmental differences in visual and auditory processing of complex sentences. Child Development, 71(4), 981–1003.

Butterworth, B. (2010). Foundational numerical capacities and the origins of dyscalculia. Trends in Cognitive Sciences, 14(12), 534–541.

Cardinaletti, A., & Volpato, F. (2015). On the comprehension and production of passive sentences and relative clauses by Italian university students with dyslexia. In E. Di Domenico, C. Hamann, & S. Matteini (Eds.), Structures, strategies and beyond: Studies in honour of Adriana Belletti (pp. 223–279). John Benjamins Publishing.

Castles, A., & Coltheart, M. (2004). Is there a causal link from phonological awareness to success in learning to read? Cognition, 91(1), 77–111.

Catts, H. W., Adlof, S. M., Hogan, T., & Weismer, S. E. (2005). Are specific language impairment and dyslexia distinct disorders? Journal of Speech, Language, and Hearing Research, 48(6), 1378–1396.

Catts, H. W., Fey, M. E., Tomblin, J. B., & Zhang, X. (2002). A longitudinal investigation of reading outcomes in children with language impairments. Journal of Speech, Language, and Hearing Research, 45, 1142–1157.

Contemori, C., & Garraffa, M. (2010). Comparison of modalities in SLI syntax: A study on the comprehension and production of non-canonical sentences. Lingua, 120, 1940–1955.

Cunnings, I. (2017). Parsing and working memory in bilingual sentence processing. Bilingualism:

Language and Cognition, 20(4), 659–678.

Danahy, K., Windsor, J., & Kohnert, K. (2007). Counting span and the identification of primary language impairment. International Journal of Language & Communication Disorders, 42(3), 349–365.

Delage, H., & Frauenfelder, U. (2019). Syntax and working memory in typically-developing children: Focus on syntactic complexity. Language, Interaction and Acquisition, 10(2), 141–176.

Delage, H., & Frauenfelder, U. (in press). Relationship between working memory and complex syntax in children with Developmental Language Disorder. Journal of Child Language.

Delage, H., Gatignol, P., & Durrleman, S. (2017). Entraîner la mémoire de travail pour améliorer la syntaxe : Une étude d’entraînement chez les enfants présentant un trouble spécifique du langage et un trouble du spectre autistique [Training working memory to improve syntax: A

(24)

training study with children with specific language impairment and autism spectrum disorder]

(Collection Unadreo). XVIIèmes rencontres internationales d’orthophonie.

De Vincenzi, M. (1991). Syntactic parsing strategies in Italian. Kluwer Academic Publishers.

Donkers, J., & Stowe, L. (2006, February 2–4). Wh-questions and the nature of D-linking: A processing perspective [Conference session]. International Conference on Linguistic Evidence, University of Tübingen, Germany.

Durrleman, S., & Delage, H. (2016). Autism spectrum disorder and specific language impairment:

Overlaps in syntactic profiles. Language Acquisition, 23(4), 361–386.

Engel de Abreu, P. M. J., Gathercole, S. E., & Martin, R. (2011). Disentangling the relationship between working memory and language: The roles of short-term storage and cognitive control. Learning and Individual Differences, 21(5), 569–574.

Fawcett, A., & Nicolson, R. I. (2008). Dyslexia and the cerebellum. In G. Reid, A. Fawcett, F.

Manis, & L. Siegel (Eds.), The SAGE handbook of dyslexia (pp. 11–29). SAGE.

Flax, J. F., Realpe-Bonilla, T., Hirsch, L. S., Brzustowicz, L. M., Bartlett, C. W., & Tallal, P.

(2003). Specific language impairment in families: Evidence for co-occurrence with reading impairments. Journal of Speech, Language, and Hearing Research, 46(3), 530–543.

Friedmann, N., Belletti, A., & Rizzi, L. (2009). Relativized relatives: Types of intervention in the acquisition of A’ dependencies. Lingua, 119, 67–88.

Friedmann, N., & Novogrodsky, R. (2011). Which questions are most difficult to understand? The comprehension of Wh questions in three subtypes of SLI. Lingua, 121(3), 367–382.

Frizelle, P., & Fletcher, P. (2015). The role of memory in processing relative clauses in children with specific language impairment. American Journal of Speech-Language Pathology, 24(1), 47–59.

Garraffa, M., & Grillo, N. (2008). Canonicity effects as a grammatical phenomenon. Journal of Neurolinguistics, 21(2), 177–197.

Gathercole, S. E., & Alloway, T. (2008). Working memory and learning: A practical guide for teachers. SAGE.

Gathercole, S. E., & Baddeley, A. (1990). Phonological memory deficits in language disordered children: Is there a causal connection? Journal of Memory and Language, 29(3), 336–360.

Gathercole, S. E., & Pickering, S. (2000). Working memory deficits in children with low achieve- ments in the national curriculum at 7 years of age. British Journal of Educational Psychology (The British Psychological Society), 70, 177–194.

Gathercole, S. E., Pickering, S. J., Ambridge, B., & Wearing, H. (2004). The structure of working memory from 4 to 15 years of age. Developmental Psychology, 40(2), 177–190.

Geary, D. C. (2004). Mathematics and learning disabilities. Journal of Learning Disabilities, 37(1), 4–15.

Guasti, M. T., Branchini, C., Vernice, M., Barbieri, L., & Arosio, F. (2015). Language disorders in children with developmental dyslexia. In S. Stavrakaki (Ed.), Specific language impairment:

Current trends in research (pp. 35–56). John Benjamins Publishing.

Håkansson, G., & Hansson, K. (2000). Comprehension and production of relative clauses: A comparison between Swedish impaired and unimpaired children. Journal of Child Language, 27(2), 313–333.

Hallahan, D. P., & Kauffman, J. N. (2003). Exceptional learners: Introduction to special education.

In P. Klouth (Ed.), Austism, autobiography, and adaptations. Teaching exceptional children (Vol. 36, pp. 42–47). Allyn & Bacon.

Hamann, C. (2006). Speculations about early syntax: The production of wh-questions by normally developing French children and French children with SLI. Catalan Journal of Linguistics, 5(1), 143–189.