Dynamic Assessment Identifies Morphosyntactic Deficits in Mono- and Bilingual Children with Developmental Language Disorder

Abstract

Dynamic Assessment (DA) is recommended for testing bilinguals as it tests the child’s learning potential and not her or his previously acquired language knowledge. Thus, it allows language difficulties to be distinguished from difficulties related to a lack of L2 exposure. This study presents the findings of DA of morphosyntax in French-speaking monolingual and bilingual children, both Typically Developing (TD) and with Developmental Language Disorder (DLD). We examined whether DA was able to distinguish TD from DLD in children, irrespective of their linguistic group (mono- versus bilingual). Morphosyntactic skills were assessed in a sample of 37 children with DLD (19 bilinguals) and 42 with TD (18 bilinguals), aged from 5 to 12. We assessed six syntactic structures: simple sentences (SVO) in present and past tense, subject relatives, accusative clitic pronouns, passives, and object relatives. We provided graduated prompts if children were not able to produce the target sentences. The results confirmed the accuracy of our morphosyntactic task to disentangle children with TD from children with DLD, regardless of their linguistic group. Moreover, cutoffs for each structure as well as for the total score, as determined via ROC curves, indicated high sensitivity and specificity for children with DLD who had a documented morphosyntactic deficit.

1. Introduction

A common concern when assessing the language capacities of bilingual children is how to distinguish difficulties due to lack of exposure to the second language (L2) from a persistent language impairment. Lexical or morphosyntactic errors can manifest similarly in these two situations, leading to a misdiagnosis of Developmental Language Disorder (DLD) in bilingual children (Chondrogianni and Marinis 2011). Misdiagnosis may indeed arise when Speech-Language Therapists (SLTs) employ standardized tests that do not take into consideration the language experience and cultural background of the child (Vender et al. 2016). Much attention has been given in recent times to alternative non-biased procedures to assess bilingual children, one of these being Dynamic Assessment (DA). In this study, we used DA to evaluate morphosyntactic skills in French-speaking mono- and bilingual children, either Typically Developing (TD) or with DLD. Whereas several studies have examined the DA of morphosyntax in English-speaking monolingual and bilingual TD and DLD children, few studies have explored the DA of morphosyntax in French.

1.1. Bilinguals’ Performance in Morphosyntax

Bilingualism is described as the capacity to communicate and/or exchange information, in more than one language (Grosjean 2018; Paradis et al. 2003). Different variables impact bilingual children’s language performance such as the age of L2 acquisition as well as the quantity and quality of language input. As an example, input quantity (as described by (Unsworth 2013) for instance) plays a significant role in L2 acquisition: an increased quantity of language input is associated with greater language proficiency in bilingual children (Unsworth 2016).

For French in particular, it has been frequently demonstrated that there is a parallel development of the first language (L1) and L2 grammars in early childhood, with no qualitative differences between monolinguals and simultaneous bilingual children (De Houwer 1995; Meisel 2017). Simultaneous bilingual children are children who acquire both their L1 and L2 before the age of three (Berken et al. 2017; Paradis et al. 2003). It has however been found that, in the production of accusative clitics, simultaneous bilinguals perform less well than their monolingual peers (Hamann and Belletti 2006; Hulk 1997). Difficulty producing accusative clitics (such as “il le prend”, “he is taking it/him”) has also been observed in sequential bilinguals, that is to say, children that learn a second language after the age of three (Berken et al. 2017; Laloi et al. 2017; Paradis 2004, 2008, 2010). More specifically, sequential bilinguals experience difficulties in the acquisition of French grammatical gender (Marchal et al. 2009), the latter being expressed in French with inflectional morphemes on determiners and adjectives1. As shown in Schönenberger (2014), determiners are frequently misused in bilingual children. Sequential bilinguals also exhibit difficulties in using past tense (Nicoladis et al. 2020; Paradis et al. 2011) and in using complex clauses such as relatives (Gil and Müller 2018; Yip and Matthews 2007) or passives (Vasilyeva et al. 2010). These deficits constitute clinical markers of DLD in French-speaking monolingual (Parisse and Maillart 2004) and bilingual children (Paradis et al. 2003).

All of the structures mentioned above are targeted in the present study, as the morphosyntactic difficulties observed in TD L2 children are similar to those observed in monolingual children with DLD (see (Fiestas and Peña 2004; Grüter 2005; Paradis 2004; Scheidnes and Tuller 2019; Scheidnes et al. 2021)). The similarity of error patterns contributes to the difficulty of diagnosing DLD in bilinguals when using a standardized morphosyntactic task in French, the language employed in this study.

1.2. Diagnosis and Assessment of Bilingual Children

DLD refers to difficulties in producing and understanding oral language causing functional impairment in everyday life (Bishop et al. 2017). DLD is often associated with poor prognosis of language improvement and the language problems associated with DLD affect multiple language domains and persist over years (Bishop et al. 2017). It has a prevalence ranging from 3 to 7% (Norbury et al. 2016), which is similar in mono- and bilingual populations (Stow and Dodd 2003; Winter 2001). In the case of bilinguals, DLD affects all of the languages spoken by the child (Kohnert et al. 2020; Paradis et al. 2011; Thordardottir et al. 2006). As mentioned above, bilingual children suffer from over- or underdiagnosis of DLD (Genesee 2008; Thordardottir 2021; Tuller et al. 2013), which can notably be related to the lack of specific norms for bilingual children in common standardized assessments (Letts 2013). Other elements that may lead to misdiagnosis in bilinguals are (1) the lack of knowledge about the characteristics of a child’s home language, especially when the variety spoken is of low status or has no written form (Pert and Letts 2006) and (2) cultural differences between clinicians and children that can lead to a misinterpretation of assessment outcomes (Isaac 2002). In order to address these numerous issues, different tools have been employed to assess bilingual children’s language more accurately, including parental questionnaires (Kašćelan et al. 2022; Paradis 2010), analysis of spontaneous language samples (Bedore et al. 2020; Rojas and Iglesias 2009), processing tests such as non-word and sentence repetition (Dos Santos and Ferré 2018; Fleckstein et al. 2018; Thordardottir et al. 2006), and DA (Hunt et al. 2022; Orellana et al. 2019), the focus of this study.

1.3. Dynamic Assessment

DA relies on assessing the learning skills of participants on a specific task, rather than measuring their static performance based on crystallized knowledge (Pena et al. 2006). It is recommended for bilingual children since it alleviates the effects of reduced language experience: indeed, the language experience is provided within the test situation (see (Hunt et al. 2022; Orellana et al. 2019) for systematic reviews on DA in bilinguals). In practice, DA can follow two different procedures: graduated prompts to assist a child in producing a target structure (e.g., phoneme, word, syntactic structure) or a pretest-teaching-post-test protocol. In the latter, improvement between the two testing times is quantified and is considered to be the consequence of the teaching phase. A measure of modifiability, corresponding to an assessment of the child’s “changeability” during a learning situation (Lidz and Peña 2009), is often used by the examiner, in addition to these procedures. DA has been shown to avoid cultural bias and is suitable for the evaluation of different linguistic domains, such as phonology, the lexicon, morphosyntax, and narration (Orellana et al. 2019). In this study, we focus on the domain of morphosyntax.

1.4. DA of Morphosyntax

Several studies have employed syntactic-based dynamic tasks with mono- and/or bilinguals and, more specifically, have investigated the learning of new morphosyntactic elements, such as morphological tense (Finestack 2018), gender (Schwob and Skoruppa, submitted), or even morphemes related to an invented aspect of an object (Roseberry and Connell 1991). For instance, Roseberry and Connell (1991) explored the teaching of a novel morpheme rule (e.g., /ono/ indicating half an object,) in bilingual Spanish–English children with TD (N = 13) and DLD (N = 13), aged from 4;5 to 6;4. They showed that after two teaching sessions of 15–20 min, participants with DLD were much poorer at learning the invented rule than those with TD: they failed to use the morpheme to name the part-object pictures provided during the probe. The authors reported a specificity of 77% and a sensitivity2 of 92% in distinguishing TD and DLD children for the invented morpheme teaching method, suggesting that their task could be useful for identifying DLD. In a similar vein, (Schwob and Skoruppa, submitted; see also Schwob 2021) used a dynamic approach to assess the learning of novel grammatical forms. Twenty-eight French-speaking monolinguals (15 TD, 13 DLD) and 21 French–Portuguese bilinguals (11 TD, 10 DLD), aged from 5 to 7, were included in their study. Bilingual children with DLD were identified by a two-step procedure: (1) the children had to have received a formal Speech and Language Therapist (SLT) diagnosis of DLD and (2) the children had to fail in at least one language domain in French and in Portuguese on standardized tests. In the task, children had to learn invented nouns, verbs, and inflections within two modalities of learning (autonomously or interactively). Children had to produce sentences with a noun and an inflected verb (with a gender morpheme at the end of it, such as “lati fakat” corresponding to: “The masculine-alien goes down and turns”) at four different moments, separated each time by a teaching phase. A grammaticality judgment task was also used to assess the children’s understanding of the use of the morphological rule. The results showed that TD children performed better than those with DLD. Unexpectedly, bilinguals and monolinguals differed on several scores with, for instance, bilinguals scoring lower than monolinguals in the grammatical judgment tasks. More importantly, the authors reported good discrimination between TD and DLD, as they found 91% specificity and 85% sensitivity on their task in diagnosing DLD within the interactive modality of learning. However, it should be noted that children with DLD (and TD) demonstrated a strong floor effect in the second and third denomination tasks, whereby participants had to produce sentences with new nouns and inflected verbs when presented with short animations of characters performing actions: participants with DLD indeed receive a mean score of 0 and their TD peers scored less than one (out of twelve stimuli). Furthermore, there was no improvement within the DA task in both mono- and bilingual DLD children between the first and last teaching phase. We might speculate that the absence of improvement by the DLD children could be due to the lack of explicit teaching in the task. Indeed, Finestack (2018) showed that children with DLD were better at learning a morphological rule based on a novel language when the rule was taught explicitly than when it was implicit.

In addition to the teaching of novel morphemes, other authors have focused on the DA of syntax, targeting simple or complex clauses. As for the early stages of grammar, Bain and Olswang (1995) employed DA with young late talkers (aged from 2;6 to 3) in order to predict which children would be able to transition from the one- to two-word stage of development. Although DA was not employed as a diagnostic tool, this study is useful for our purposes as it shows that hierarchical graduated prompts can enable children to move from one grammatical stage to another. The results confirmed the validity of the cue hierarchy since children were able to produce more target two-word utterances when more supportive cues were provided. One of the verbal cues provided was a direct model aimed at eliciting spontaneous imitation (“see, the dog is walking; what is it doing?”—expected answer: “dog walking”). Moreover, children who produced fewer target structures during the treatment phase produced less changes during the 9-week study period than those who produced more target sentences during the treatment period.

As for more advanced stages of grammatical development, Hasson and colleagues (Hasson et al. 2012, 2013) have developed two DA tasks of syntax. The first one was based on the Sentence Assembly subtest of the CELF-3 (Semel et al. 1995) and was designed to test children aged from 8 to 10 years. The task involved children rearranging groups of printed words to create grammatical structures. The children were required to produce two different sentences based on the same stimuli (for instance, words such as “drinking—and—is—Mum—Dad—eating—is” are presented to the child who should create two distinct sentences, namely “Mum is eating and Dad is drinking” and “Dad is eating and Mum is drinking”). The prompts were graded from a general metacognitive direction (“Do you know what you have to do?”) to item specific feedback and instruction (modeling) and were based on methods previously used by Resing (2013). The authors administered this task over two sessions and found that gains on the task, from first to second session, were correlated with scores on a standardized test of language. They concluded that their DA task was therefore suitable for use with DLD children. In a second DA task, a subtest of the DAPPLE (Dynamic Assessment of Preschoolers’ Proficiency in Learning English), Hasson et al. (2013) targeted younger children, aged from 3 to 5. This approach employed a pretest-teaching-post-test procedure. At pretest, the children had to describe in one sentence what was happening in a picture (i.e., The boy is eating the banana). During the teaching phase, graduated prompting was delivered by the examiner, in the form of elicitation (i.e., asking questions on the actions) or modeling (i.e., providing the correct structure to be repeated), similar to the prompts used by Bain and Olswang (1995). Finally, children had to describe the same pictures, with no prompting provided in the post-test. Twelve syntactic structures (e.g., declarative sentences with embedded clauses) were targeted and items were presented in order of increasing difficulty and/or increasing length/number of items in the sentence, for each grammatical structure. Hasson et al. (2013) administered this task to 25 children with a mean age of 4;2: 12 bilingual caseload children (=children receiving SLT) and 13 TD bilinguals. Both groups (TD and caseload) performed higher in the post-test, with TD bilingual children scoring significantly better than the caseload group. Specifically, both groups used more subject–verb structures in the post-test than in the pretest, which resulted from a decrease in verb–object (with subject omission) and an increase in subject–verb structures in the caseload group, and from a higher production of conjoined clauses in the TD group. There were also fewer subject–verb–object structures and more four-element subject–verb–object–adverbial structures in the TD group in the post-test.

As described above, the morphosyntax can be assessed using a dynamic approach by providing children with graduated prompts if they cannot produce the target structure. However, no work to date has been conducted in French; only one study has used novel words with French-speaking children to teach them invented gender morphological markers (Schwob and Skoruppa, submitted).

1.5. The Current Study

As previously mentioned, DA evaluates the language learning potential rather than crystalized language knowledge, which makes this kind of assessment particularly useful for assessing the language capacities of bilingual children. These children may indeed suffer from a lack of exposure to the language employed in the assessment that can lead to errors similar to those produced by children with DLD. Only a handful of studies have demonstrated the utility of the DA of morphosyntax for distinguishing between TD and DLD in bilinguals (Bain and Olswang 1995; Hasson et al. 2012, 2013), of which only one has been conducted in French (Schwob and Skoruppa, submitted). As indicated above, the latter study was subject to strong floor effects. Considering these elements, we have developed a new DA task of morphosyntactic skills, with hierarchical prompts designed to help children produce various target structures.

Thus, the central goal of the current study is to explore the accuracy of a new DA task of morphosyntax in diagnosing DLD in a population of French-speaking mono- and bilingual children. We have two main questions:

• Which variables affect children’s performance on our syntactic DA task?

We hypothesize that bilinguals will perform equally to monolinguals and that TD children will score higher than children with DLD; in other words, we expect an impact of the diagnostic group and no effect of linguistic group on the DA scores. In addition, we hypothesize that age and nonverbal reasoning will also be linked to children’s performance. Indeed, our dynamic task relies on learning skills since children have to be able to recognize and exploit the presence of analogous target structures that are provided both within (via syntactic priming) and between trials (with three items for each target structure). Analogous reasoning is known to be predicted by nonverbal reasoning (Griffiths et al. 2022).

2.

Does our morphosyntactic DA task diagnose children with DLD?

In order to answer this question, we will conduct specificity and sensitivity analyses to determine the clinical utility of the morphosyntactic DA task in distinguishing children with TD and DLD.

2. Materials and Methods

2.1. Participants

A total of 79 French-speaking participants, aged from 5;1 to 11;9 (M = 7;9), were recruited in Genevan school districts and in France. They were divided into four groups: monolinguals with TD or DLD and bilinguals with TD or DLD (see

Table 1. Descriptive data on participants.

Diagnostic Group	Linguistic Group	N	Gender	Mean Age Years; Month (SD)	Age Range	Non-Verbal Reasoning Raw Score (SD)	Non-Verbal Reasoning z Score (SD)	Morphosyntactic Production Raw Score (SD)	Morphosyntactic Production z Score (SD)
TD	Mono	24	14f	7;9 (1;3)	6;0–10;9	26.8 (6.15)	0.09 (0.89)	17.4 (7.61)	0.87 (1.60)
	Bi, 5 seq.	18	10f	7;1 (1;5)	5;1–9;7	23.0 (7.32)	−0.16 (1.07)	14.8 (7.73)	0.79 (1.93)
DLD	Mono	18	10f	7;7 (0;8)	6;6–11;9	22.6 (5.92)	−0.53 (1.16)	7.33 (5.49)	−1.59 (1.50)
	Bi, 9 seq.	19	7f	8;3 (1;6)	6;0–11;9	26.6 (4.11)	−0.10 (0.93)	7.83 (3.81)	−2.08 (1.14)

Mono = monolinguals; Bi = bilinguals; seq. = sequential bilinguals; f = females; SD = standard deviation.

). The diagnostic group refers to the distinction TD versus DLD. The children with DLD, either mono- or bilingual, received a diagnosis of language impairment by a certified SLT. More specifically, children with DLD had been tested by clinicians in Switzerland or France, with common French standardized language tests, such as Exalang 5−8 (Thibault et al. 2003) or Evaleo 6-15 (Launay et al. 2018). The clinicians applied the usual inclusionary and exclusionary criteria to diagnose DLD, which included children having scores of at least two SDs below the norm on static language assessments (Classification Statistique Internationale des Maladies et des Problèmes de Santé Connexes). We ensured via a parent questionnaire that the children did not present any known differentiation condition (as defined by the CATALISE group, Bishop et al. 2017) such as brain injury, aphasia, cerebral palsy, sensorineural hearing loss, or autism spectrum disorder. We also verified that all children had scores that did not fall below two SDs of the mean on the Raven’s progressive matrices (Raven et al. 1998), thus excluding any risk of intellectual disability, and that TD children had normal levels of language (i.e., they did not receive language therapy) and were functioning without difficulties in their age-appropriate classroom, according to their parents. Finally, a parental questionnaire allowed us to ascertain whether the child was bilingual and whether French was her/his dominant language. We considered as bilingual any child who could speak another language apart from French more than 30% of her/his time (with French always being the language of instruction at school), following the criteria of Paradis (2010). We also noted whether the child was a sequential or simultaneous bilingual, as well as the L1 spoken by the child. The first languages were distributed as follows: Portuguese (7), English (7), Arabic (5), Italian (4), Greek (3), Spanish (3), German (2), Russian (2), Swiss German (2), Bosnian (1), Bulgarian (1), Kurdish (1), Serbian (1), Turkish (1), and Ukrainian (1). Seven children spoke more than two languages and appear therefore twice in the enumeration of L1s. Moreover, two children were classified as bilingual even though they were not fluent in their L1, according to their parents. They were indeed exposed to their L1 (Arabic in both cases) from birth, more than 30% of the time.

A morphosyntactic task (BILO 3C, Khomsi et al. 2007) was administered to all participants before the DA task and all instructions were in French. The participants did not differ significantly by age (z = 0.87, p = 0.39) nor by non-verbal reasoning, as assessed by the Raven’s matrices (z = −0.50 p = 0.62). TD mono- and bilingual participants did not differ by the morphosyntactic production raw score (z = 0.23, p = 0.82) and neither did DLD mono- and bilingual participants (z = 0.90, p = 0.37). As expected, children with DLD obtained significantly lower morphosyntactic production scores than their TD peers (z = −3.97, p < 0.001) when raw scores or z scores were compared.

2.2. General Procedure

All testing was conducted by master students in speech-language therapy or by the first author. The children were tested at school or in their homes (N = 47) or online via the Zoom platform (N = 36), the latter due to the coronavirus COVID-19 pandemic3. The parents provided informed, written consent, allowing their child to participate in the study. The study was conducted in accordance with the Declaration of Helsinki and approval for the research was obtained from the Ethics Committee of the Faculty of Psychology and Educational Sciences of the University of Geneva.

Our new task was designed to assess six morphosyntactic structures (see

Table 2. Target grammatical structure.

Simple Grammatical Structures	Complex Grammatical Structures
Subject–Verb–Object (SVO)	Sentence with a subject relative (SR)
Ex: La grand-mère tire le chien	Ex: Le garçon qui prend la pomme
“The grandmother pulls the dog”	“The boy who is taking the apple”
SVO-sentence using a past tense	Sentence with an accusative clitic pronoun
Ex: Le père a couvert le garçon	Ex: La fille le pousse
“The father covered the boy”	“The girl pushes him”
	Passive sentence
	Ex: La balle est lancée par le garçon
	“The ball is thrown by the boy”
	Sentence with an object relative (OR)
	Ex: Le garçon que la fille porte
	“The boy that the girl carries”

), which pose difficulty for children with DLD in French (see Maillart et al. 2012). The task is partly based on a clinical protocol developed by Leroy (2019) for a clinical use only. Applying a DA approach, graduated prompts were provided in a hierarchal order, from the least to the most directive. If the child did not produce the target structure after the first question (e.g., “Which boy is it?”), the examiner provided her/him an initial cue. If the child still failed to produce the target structure, a second cue was provided, and so on, until the last condition, in which the child was requested to repeat the target (e.g., “It is the boy who takes the apple”, see

Table 3. An example of graduated prompts and scoring for a subject relative.

v	Points Accorded	Description	Example
		Là le garçon prend la pomme. Ici le garçon mange la pomme. “Here, the boy is taking the apple. Here, the boy is eating the apple.”
No help needed to produce the target	6	Alors là, c’est quel garçon? “So here, which boy is it?” Expected answer: C’est le garçon qui prend la pomme. “It is the boy who is taking the apple.”
Visual priming	5	Red arrow pointing to the agent (or the patient for passives and object relatives).		“Which boy is it?” Expected answer: “It is the boy who is taking the apple.”
Gradual morphosyntactic priming	4	A parallel example with different characters is provided by the examiner, followed by a return to the initial question (with the initial picture).		“Which mother is it? It is the mum who is taking the bowl. Now which boy is it?” Expected answer: “It is the boy who is taking the apple.”
	3	Another example is provided, with the same character as before asking again the initial question (with the initial picture).		“What boy is it? It is the boy who is taking the banana. Now, which boy is it?” Expected answer: “It is the boy who is taking the apple.”
	2	Gradual completion of the correct sentence provided by the examiner.	It…. It is …. It is the boy….
	1.5	Only one word is left for the child to say.	It is the boy who is taking the…
Modeling	1	Repetition of the target sentence.	It is the boy who is taking the apple.
	0	Failure to repeat the target sentence.

). Each target structure was assessed by three items, two of them being a reversible structure (=subject and object can both be a grammatical agent or patient, as in the passive structure: “The girl is carried by the boy”) and one being non-reversible (=subject and object cannot be reversed, as in “the ball is thrown by the boy”). Before the elicitation task, the children were first familiarized with the seven verbs used in the task (to wash, to pull, to push, to take, to cover, to carry, and to throw) by hearing the verb spoken and by seeing pictures of Playmobile characters in action (ex: you see, this is “to wash”), as in Hasson et al. (2012). The children were then asked to produce the target morphosyntactic structures by answering specific questions (see details in Appendix A,

Table A1. An example of one of the three items, for each targeted structure.

French Instructions			Translation in English
Ça c’est « Laver ». Ça c’est « Tirer ». Ça c’est « Pousser ». Ça c’est « Prendre ». Ça c’est « Couvrir ». Ça c’est « Porter ». Ça c’est « Lancer ».			“This is ‘wash’. This is ‘pull’. This is ‘push’. This is ‘take’. This is ‘cover’. This is ‘carry’. This is ‘throw’.”
Maintenant, je vais te montrer des images avec des personnages et tu dois me dire ce que tu vois en faisant une phrase.			“Now, I will show you some pictures with characters and you tell me what you see by making a sentence.”
Subject–Verb–Object (SVO)-sentence
Le papa pousse le cochon. “The father is pushing the pig”	6	Que fait le papa?	“What is the father doing?”
	5	Que fait le papa?	“What is the father doing?”
	4	Que fait la maman? La maman pousse la chèvre. Et là, que fait le papa?	“What is the mother doing? The mother is pushing the goat. So here, what is the father doing?”
	3	Que fait le papa? Le papa pousse le mouton. Et là, que fait le papa?	“What is the father doing? The father is pushing the sheep. So here, what is the father doing?”
	2	Que fait le papa? Le papa… Le papa pousse	“What is the father doing? The father…” “The father is pushing…”
	1.5	Le papa pousse le	“The father is pushing the…”
	1	Le papa pousse le cochon. Répète après moi	“The father is pushing the pig. Repeat after me.”
SVO-sentence using past tense
La mamie a couvert/couvrait le garçon. “The grandmother covered the boy.”	6	Là, la mamie couvre le garçon. → Alors, tu te souviens, qu’est-ce qu’il s’est passé sur la photo avec la mamie?	‘Here, the grandmother is covering the boy. → Do you remember what happened on the picture with the grandmother?’
	5	Là, la mamie couvre le garçon. → Alors, tu te souviens, qu’est-ce qu’il s’est passé sur la photo avec la mamie?	“Here, the grandmother is covering the boy. → Do you remember what happened on the picture with the grandmother?”
	4	Là, le papi couvre la fille. Alors tu te souviens, qu’est-ce qu’il s’est passé sur la photo avec le papi? Le papi a couvert la fille. Là, la mamie couvre le garçon. → Alors, tu te souviens, qu’est-ce qu’il s’est passé sur la photo avec la mamie?	“Here, the grandfather is covering the girl. Do you remember what happened on the picture with the grandfather? The grandfather covered the girl.” “Here, the grandmother is covering the boy. → Do you remember what happened on the picture with the grandmother?”
	3	Là, la mamie couvre le papa. Alors tu te souviens, qu’est-ce qu’il s’est passé sur la photo avec la mamie? La mamie a couvert le papa. Là, la mamie couvre le garçon. → Alors, tu te souviens, qu’est-ce qu’il s’est passé sur la photo avec la mamie?	“Here, the grandmother is covering the father. → Do you remember what happened on the picture with the grandmother? The grandmother covered the father.” “Here, the grandmother is covering the boy. → Do you remember what happened on the picture with the grandmother?”
	2	Qu’est-ce qu’il s’est passé sur la photo avec la mamie? La mamie… La mamie a couvert…	“Do you remember what happened on the picture with the grandmother? The grandmother…” “The grandmother covered…”
	1.5	La mamie a couvert le…	“The grandmother covered the…”
	1	La mamie a couvert le garçon. Répète après moi.	“The grandmother covered the boy. Repeat after me.”
Subject relative
La maman qui couvre la fille. “The mother who is covering the girl.”	6	Ici, la maman couvre la fille. → Là, la fille couvre la maman. → Alors là, c’est quelle maman?	“Here, the mother is covering the girl. → Here, the girl is covering the mother. → So here, which mother is it?”
	5	Alors celui-là, c’est quelle maman?	‘So here, which mother is it?’
	4	C’est quel papa? C’est le papa qui couvre le garçon. → Alors celui-là, c’est quelle maman?	“Which father is it? It is the father who is covering the boy. → So here, which mother is it?”
	3	C’est quelle maman? C’est la maman qui couvre la mamie. → Alors celle-là, c’est quelle maman?	“Which mother is it? It is the mother who is covering the grandmother. → So here, which mother is it?”
	2	C’est quelle maman? C’est la maman… C’est la maman qui… C’est la maman qui couvre…	“Here, which mother is it? It is… It is the mother who… It is the mother who is covering…”
	1.5	C’est la maman qui couvre la…	“It is the mother who is covering the…”
	1	C’est la maman qui couvre la fille. Répète après moi.	“It is the mother who is covering the girl. Repeat after me.”
Sentence with an accusative clitic pronoun
La fille le pousse. “The girl is pushing him.”	6	Que fait la fille avec le cheval?	“What is the girl doing with the horse?”
	5	Que fait la fille avec le cheval?	“What is the girl doing with the horse?”
	4	Que fait le papa avec les chiens? Il les pousse. → Alors là, que fait la fille avec le cheval?	“What is the father doing with the dogs? He pushes them. So here, what is the girl doing with the horse?’’
	3	Que fait la fille avec la chèvre? Elle la pousse. → Alors là, que fait la fille avec le cheval?	“What is the girl doing with the goat? She pushes it-feminine marker. So here, what is the girl doing with the horse?”
	2	Que fait la fille avec le cheval? Elle…	“What is the girl doing with the horse? The girl…”
	1.5	Elle le…	“The girl pushes…”
	1	Elle le pousse. Répète après moi.	“The girl pushes him. Repeat after me.”
Passive
La fille est portée par le papi. / La fille se fait porter par le papi. “The girl is carried by the grandfather.”	6	Qu’est-ce qui arrive à la fille?	“What happens to the girl?”
	5	Qu’est-ce qui arrive à la fille?	“What happens to the girl?”
	4	Qu’est-ce qui arrive au papa? Le papa est porté par la maman. → Et là, qu’est-ce qui arrive à la fille?	“What happens to the father? The father is carried by the mother. → And here, what happens to the girl?”
	3	Qu’est-ce qui arrive à la fille? La fille est portée par le garçon. → Et là, qu’est-ce qui arrive à la fille?	“What happens to the girl? The girl is carried by the boy. → And here, what happens to the girl?”
	2	Qu’est-ce qui arrive à la fille? La fille La fille est portée… La fille est portée par…	“What happens to the girl? The girl… The girl is carried… The girl is carried by…”
	1.5	La fille est portée par le …	“The girl is carried by the…”
	1	La fille est portée par le papi. Répète après moi.	“The girl is carried by the grandfather. Repeat after me.”
Object relative
La fille que le garçon porte. “It is the girl that the boy carries.”	6	Ici, le garçon porte la fille. → Là, la fille porte le garçon. → Alors celle-là, c’est quelle fille?	“Here, the boy carries the girl. → There, the girl carries the boy. → So here, which girl is it?”
	5	Alors celle-là, c’est quelle fille?	“So here, which girl is it?”
	4	C’est quel papi? C’est le papi que la mamie porte. → Alors celle-là, c’est quelle fille?	“Which grandfather is it? It is the grandfather that the grandmother carries. → So here, which girl is it?”
	3	C’est quelle fille? C’est la fille que le papa porte. → Alors celle-là, c’est quelle fille?	“Which girl is it? It is the girl that the father carries. → So here, which girl is it?”
	2	C’est quelle fille? C’est la fille… C’est la fille que…	“Which girl is it? It is the girl… It is the girl that…”
	1.5	C’est la fille que le garçon…	“It is the girl that the boy…”
	1	C’est la fille que le garçon porte. Répète après moi.	“It is the girl that the boy carries. Repeat after me.”

). Two of the seven target structures were simple syntactic structures, that is, sentences that followed a canonical order (subject–verb–object), without any embedding or phrasal movement (Delage and Frauenfelder 2020); they contained a present or past tense verb. The four other structures were complex ones: i. with embedding and movement (but a canonical order of constituents and no sensitivity to intervention effects4), as in a subject relative; ii. with syntactic movement, where movement places the accusative clitic in a noncanonical, preverbal position; iii. with another type of object movement found in passives (see Stanford and Delage 2021); and iv. with both embedding and phrasal movement, as in an object relative involving intervention effects (due to the grammatical similarity of the two NPs, see (Bentea and Durrleman 2019)). The six target structures are provided in Table 2.

For all target structures, the Determiner Phrase (DP) varied in gender (the two reversible DP items being masculine and feminine) but did not vary in number (all items were singular). Finally, each target sentence had the same mean number of syllables (see Appendix A,

Table A2. Syntactic items.

	Grammatical Structure Targeted	Reversibility (Y = Yes; N = No)	Item	Mean Number of Syllables (SD)
Simple grammatical structures	Subject–Verb–Object (SVO)	N	Le garçon lance la pomme “The boy throws the apple”	6.33 (0.58)
		Y	Le papa pousse le cochon “The father pushes the pig”
		Y	La mamie tire le chien “The grandmother pulls the dog”
	SVO-sentence using past tense	N	Le papa a pris le seau “The father took the bucket”	8.33 (1.15)
		Y	La mamie a couvert le garçon “The grandmother covered the boy”
		Y	La maman a lavé le papi “The mother washed the grandfather”
Complex grammatical structures	Subject relative	N	Le garçon qui prend la pomme “The boy who takes the apple”	7.33 (0.56)
		Y	La maman qui couvre la fille “The mother who covers the girl”
		Y	Le cheval qui porte la mamie “The horse that carries the grandmother”
	Sentence with an accusative clitic pronoun	N	Il les prend “He takes them”	3
		Y	Il la couvre “He covers her”
		Y	Elle le pousse “She pushes him”
	Passive	N	Le ballon est lancé par le garçon “The ball is thrown by the boy”	9.33 (0.57)
		Y	La fille est portée par le papi “The girl is carried by the grandfather”
		Y	Le chien est tiré par la maman “The dog is pulled by the mother”
	Object relative	N	Le doudou que le papa lance “The stuffed toy that the father throws”	7.33 (0.58)
		Y	La fille que le garçon porte “The girl that the boy carries”
		Y	Le garçon que la fille lave “The boy that the girl washes”

). The answers were considered correct if they contained the targeted grammatical structure (as in Finestack 2018), even if there was a lexical substitution (ex: “nanny” instead of “grand-mother”), or a gender error (ex: le-masc. maman “mummy” instead of la-fem. maman “mummy”)5. If the child’s answer was incorrect or incomplete, if it did not include the target structure (with for example an omission of the tense marker or a determiner phrase for the accusative clitic), graduated prompts were provided, as previously explained, and the points were attributed according to the number of cues needed to produce the target. If an error was created after the first prompt, the examiner continued to provide hierarchical prompting, until the child produced the target structure. Table 3 displays the scoring, detailing the prompts necessary for the child to generate the target morphosyntactic structure. The children could be allocated 18 points maximum per structure, with three items per structure, that is, 108 points in total on the DA task. As children were tested either by trained master students or by the first author, all the answers leading to a prompt were checked by the first author, in order to ensure that the appropriate prompt was provided. The responses at all stages of the procedure were verified to ensure that the correct number of points were attributed to them.

3. Results

The data analyses had two aims: (1) to investigate the impact of age, and linguistic and diagnostic groups as well as nonverbal reasoning on dynamic scores; (2) to calculate the diagnostic power of our syntactic DA task.

The parametric tests were used, namely linear regression, partial correlation, and F tests. The ROC analyses were conducted to determine the optimal cutoffs on the basis of the sensitivity and specificity results. The analyses were performed using R statistical software (R Core Team 2020).

3.1. Preliminary Analysis

In order to illustrate diagnostic group differences, the morphosyntactic raw scores (to remind the reader, the higher the number of cues, the lower the score) were analyzed structure by structure. Our group comparison showed that children with DLD needed significantly more cues in order to produce almost all target structures than TD children: for SVO-sentences using a past tense (z = −3.70, p < 0.001, d = 0.93), subject relatives (‘SR’, z = −2.83, p = 0.005, d = 0.68), sentences with an accusative clitic pronoun (z = −4.95, p < 0.001, d = 1.35), passives (z = −4.06, p < 0.001, d = 1.04), and object relatives (‘OR’, z = −4.09, p < 0.001, d = 1.04), as well as for the total score, collapsing all structures (z = 4.09, p < 0.001, d = 1.38). As for the very basic structure SVO, no difference was found between the two groups (z = −0.29, p = 0.77). These results are illustrated in Figure 1.

3.2. Which Variables Affect Syntactic DA Score?

In order to look for variables that would uniquely contribute to accounting for the total DA score, we computed a multiple linear regression. In this way, we first verified the contribution of each variable in explaining the total syntactic DA score (F(5,73) = 31.09; p < 0.001), these variables being age, nonverbal reasoning, and linguistic and diagnostic groups, as well as the interaction between the linguistic and diagnostic groups. The results are displayed in

Table 4. Linear regression of total DA score using age, nonverbal reasoning z score, and diagnostic and linguistic groups and the interaction of these two as independent variables.

	Estimate	Standard Errors	T	p
Intercept	32.98	9.55	3.46	<0.001
Nonverbal reasoning z score	8.75	1.48	5.92	<0.001
Age	0.65	0.10	6.82	<0.001
Diagnostic group	−20.74	4.12	−5.04	<0.001
Linguistic group	−2.00	4.13	0.48	0.63
Diagnostic×Linguistic group	−10.54	6.17	−1.71	0.09

p-values in bold indicate significant contributions.

. This first analysis highlights the contribution of the diagnostic group, age, and nonverbal reasoning on the total dynamic score, with no effect of the linguistic group (i.e., bilingualism) nor of the interaction between the linguistic and diagnostic groups. In order to ensure that bilingualism had no effect on the dynamic score, we computed a second linear regression with age, nonverbal reasoning, and diagnostic group in the model but without adding linguistic group to the model (and consequently neither the interaction between the diagnostic and linguistic groups). This model (F(3,75) = 49.13; p < 0.001), the results of which are displayed in

Table 5. Linear regression of total DA score using age, nonverbal reasoning z score, and diagnostic group as independent variables.

Total DA Score	Estimate	Standard Errors	T	p
Intercept	38.03	8.62	4.41	<0.001
Nonverbal reasoning z score	8.18	1.46	5.59	<0.001
Age	0.61	0.09	6.60	<0.001
Diagnostic group	−25.97	2.96	−8.77	<0.001

p-values in bold indicate significant contributions.

, was finally compared to the more complete model (F(2,73) = 2.02, p = 0.14): the difference between the two models was not significant. In other words, since the model without the linguistic group was not different from the model that included the linguistic group, we conclude that bilingualism does not play a role in accounting for the total morphosyntactic DA score.

In order to investigate each target structure independently, we repeated the three step analyses on each structure. The models with and without the linguistic group were systematically compared and no difference was found between the two models for any structure (see results in Appendix A,

Table A3. Linear regression of each structure score using age, nonverbal reasoning z score, and diagnostic and linguistic groups and the interaction of these two as independent variables.

	SVO			SVO-Sentence Using a Past Tense			Subject Relative (SR)			Sentence with an Accusative Clitic Pronoun			Passive			Object Relative (OR)
	E	t	p	E	t	p	E	t	p	E	t	p	E	t	p	E	t	p
Intercept	15.54	13.89	<0.001	2.65	1.00	0.32	4.66	1.81	0.07	−1.30	−0.47	0.64	−0.57	−0.25	0.81	−0.36	−0.16	0.88
NVR z score	0.17	0.92	0.36	1.03	2.41	0.02	1.23	2.96	0.004	0.82	1.82	0.07	1.53	4.13	<0.001	1.43	3.76	<0.001
Age	0.02	1.51	0.14	0.11	3.97	<0.001	0.09	3.49	<0.001	0.10	3.43	0.001	0.13	5.22	<0.001	0.10	3.88	<0.001
Diagnostic group	−0.20	−0.40	0.69	2.29	1.93	0.06	0.93	0.80	0.43	5.31	4.22	<0.001	3.79	3.66	<0.001	2.64	2.50	0.01
Linguistic group	0.10	0.19	0.85	−2.07	−1.66	0.10	−1.87	−1.54	0.13	−0.44	−0.36	0.74	−0.26	−0.24	0.81	−1.68	−1.51	0.13
Diagnostic×Linguistic group	0.48	0.63	0.53	2.79	1.57	0.12	2.72	1.56	0.12	0.87	0.46	0.64	0.31	0.20	0.84	2.27	1.43	0.16
Comparisons of models with and without Linguistic group	F(1-73) = 0.40; p = 0.53			F(1-73) = 2.16; p = 0.12			F(1-73) = 2.45; p = 0.12			F(1-73) = 0.22; p = 0.64			F(1-73) = 0.04; p = 0.84			F(1-73) = 2.06; p = 0.16

E = Estimate; NVR z score = nonverbal reasoning z score; p-values in bold indicate significant contributions.

). Thus, the linguistic group as an independent variable did not influence our dependent measures (DA score of each structure). As the comparison of the two models did not reveal any significant differences, we decided to only display the results of the most complete model for each structure, that is to say, the model including the linguistic group (see Appendix A, Table A3). Age explains a significant part of the variance in scores for all structures, except for the simplest one, namely SVO. Moreover, the diagnostic group (TD versus DLD) is a significant variable in sentences with an accusative clitic pronoun, passive, and OR. A marginal contribution of the diagnostic group was also found on SVO-sentences using past tense, but there was no effect of the diagnostic group on SVO and SR. Finally, nonverbal reasoning emerged as significant for most structures: SVO-sentences using a past tense, SR, passive, and OR.

3.3. Accuracy in DLD Diagnosis

As observed above, the diagnostic group was a significant variable for most of the complex structures. In order to check for diagnostic accuracy, we first computed a stepwise multiple regression to select the variables that uniquely contributed to classification accuracy. The results are displayed in

Table 6. Stepwise logistic regression: DLD diagnosis (dependent variable), and Total DA morphosyntactic score, nonverbal reasoning z score, and age (independent variables).

DLD (N = 37)
Variables entered a	β	z	p
1. Total DA morphosyntactic score	−0.17	−4.47	<0.001
2. Age	0.11	3.51	<0.001
3. Nonverbal reasoning z score	1.25	2.57	0.01

^a Excluded variables in final model: Linguistic group. p-values in bold indicate significant contributions.

and highlight the contribution (p < 0.001) of both the DA score and age in classifying the children with TD versus DLD. It should be noted that linguistic group did not affect this classification.

In order to determine the cutoffs that resulted in the discrimination of TD and DLD children using our syntactic DA task, we considered each score obtained on the DA task and age. We note that the nonverbal reasoning score accounted for DLD diagnosis when considering the DA total score, as shown in Table 6. However, it was not the case when we computed stepwise logistic regression and considered the score of each structure individually. When we did so, nonverbal reasoning was always excluded from the model as was the case for the linguistic group and only the structure score and age emerged as significant in the logistic regression models. Therefore, we only present cutoffs that consider score and age.

Table 7. Classification accuracy of DLD participants versus TD children by cutoff points.

Cutoffs	Sensitivity	Specificity	Accuracy	AUC	LR+	LR−
Total DA Morphosyntactic Score
Total DA score < 17.23 + 0.65 × Age	81%	93%	87%	0.91	11.57	0.20
SVO
SVO < −80.71 + 1.24 × Age	95%	29%	59%	0.55	1.34	0.17
Past tense
Past < −3.12 + 0.15 × Age	54%	88%	72%	0.76	4.50	0.52
Subject relatives
SR < −1.94 + 0.16 × Age	68%	67%	67%	0.68	2.06	0.48
Sentence with an accusative clitic pronoun
Clitic < −4.49 + 0.94 × Age	62%	93%	78%	0.86	8.86	0.41
Passives
Passive < −3.44 + 0.15 × Age	54%	100%	78%	0.80	0	0.46
Object relatives
OR < −2.44 + 0.12 × Age	68%	88%	78%	0.80	5.67	0.36

displays sensitivity, specificity, and accuracy rates corresponding to each cutoff as well as positive likelihood ratios (LR+) and negative likelihood ratios (LR−)6 for total DA score and separate structure scores.

LR+ and LR− have been calculated (as in Armon-Lotem et al. 2015), given the following formula, mentioned in Dollaghan (2004):

$$\mathrm{LR}+=\frac{sensitivity}{1-specificity}\mathrm{and}\mathrm{LR}-=\frac{specificity}{1-sensitivity}$$

The accuracy has been calculated by using the following formula (see Dollaghan 2004):

$$\mathrm{Accuracy}=\frac{True Positive+True Negative}{Positive+Negative}$$

For each structure, the negative likelihood ratios are all above 0.17 and can be considered as “diagnostically uninformative” in differentiating TD from children with DLD, following Dollaghan’s classification (Dollaghan 2007). It seems to us that these discouraging results could be due to the initial identification of children with DLD. To recall, children with DLD, either mono- or bilingual, received a diagnosis of language impairment by a certified SLT to be included in the study. However, we would like to point out that, out of the 37 children with DLD, only 23 of them obtained a pathological score on the standardized morphosyntactic task (BILO 3C, Khomsi et al. 2007), that is to say, a z score below −1.65 SD (i.e., below the fifth percentile, which is the common cutoff used in French speaking countries to identify a DLD, Tuller et al. 2017). This led us to form a subgroup in our DLD sample, the latter being DLD children with a documented morphosyntactic impairment.

3.4. Accuracy in the Diagnosis of DLD Children Displaying a Documented Morphosyntactic Impairment

As illustrated in Figure 2, there is an overlap between the scores of TD and DLD children on the standardized test of syntax, which could indicate that some DLD participants have morphosyntactic abilities similar to those of TD children. Indeed, as explained above, when looking at the individual scores obtained by the DLD children on the standardized morphosyntactic task (BILO 3C, Khomsi et al. 2007), 13 children with DLD had z scores above −1.65 SD (based on normative data). These relatively intact morphosyntactic scores may be due to the fact that the clinical diagnosis of DLD is based on deficits in at least two language domains, the latter not necessarily including the morphosyntax. Therefore, we decided to conduct the analysis presented above on a restricted sample of children with and without morphosyntactic impairment, in other words, in children that obtained a z score above or below −1.65 SD, on the BILO task. Thus, we considered as “morphosyntactically impaired” the children with DLD who obtained a z score below −1.65 SD on the BILO task (N = 21).

In order to check for diagnostic accuracy, we first computed a stepwise multiple regression to select the variables that uniquely contributed to classification accuracy (TD versus morphosyntactically impaired DLD children). The results are displayed in

Table 8. Stepwise logistic regression: classification of morphosyntactic impairment (dependent variable), using total DA morphosyntactic score and age as independent variables.

Morphosyntax Impairment (N = 23)
Variables entered a	β	z	p
1. Total DA morphosyntactic score	0.14	4.04	<0.001
2. Age	−0.13	−3.48	<0.001

^a Excluded variables in final model: linguistic group, nonverbal reasoning z score. p-values in bold indicate significant contributions.

7; they highlight the contribution of both the DA score and age in classifying TD versus DLD children with morphosyntactic impairment.

As age and the total score on the morphosyntactic DA task were significant, we generated ROC curves using these two variables. We report in

Table 9. Classification accuracy of DLD participants with morphosyntactic deficiency versus TD children by cutoff points.

Cutoffs	Sensitivity	Specificity	Accuracy	AUC	LR+	LR−
Total DA Morphosyntactic Score
Total DA score < −22.7 + 0.94 × Age	76%	95%	89%	0.95	15.20	0.25
SVO
SVO < −10.50 + 0.23×ge	28%	96%	72%	0.70	7.00	0.75
Past tense
Past < −18.12 + 0.28 × Age	52%	95%	83%	0.80	10.40	0.51
Subject relatives
SR < −11.84 + 0.22 × Age	48%	95%	82%	0.85	9.60	0.55
Sentence with an accusative clitic pronoun
Clitic < −14.34 + 0.22 × Age	76%	95%	89%	0.90	15.20	0.25
Passives
Passive < −10.42 + 0.22 × Age	86%	93%	91%	0.93	12.29	0.15
Object relatives
OR < −12.40 + 0.20 × Age	81%	93%	89%	0.90	11.57	0.20

the probability of receiving a diagnosis of a morphosyntactic impairment (e.g., a raw score under −1.65 SD on the standardized morphosyntactic task associated with a formal DLD diagnosis), using different cutoffs. Specifically, we present cutoffs using a similar method to the one in the previous section, which was determined by using logistic regression models. The cutoffs are presented with respect to age, with each corresponding classification accuracy, true positive rate (sensitivity), false-positive rate (specificity), and area under the curve (AUC) as well as the likelihood ratios for a positive test (LR+) and likelihood ratios for a negative test (LR−). We present cutoffs corresponding to each structure for the highest accuracy score.

The children diagnosed with DLD by SLTS but without documented syntactic deficiency were removed from these analyses for the reasons explained above. The results showed that the scores on the complex syntactic structures (namely subject relatives, accusative clitics, passives, and object relatives), as well as the total dynamic score, distinguished children with TD and those with DLD with good accuracy (all scores over 80%, as mentioned in Swets et al. 2000), as well as high specificity.

To help the reader understand Table 9, we provide the following examples: we consider a child that we tested (aged 7;8, that is 92 months), who obtained a total score of 47.5 on the morphosyntactic DA task. Given the cutoffs and formula we indicated in Table 9, we would consider this child as DLD and Morphosyntax deficient if her/his score falls under 80.81 (=−5.67 + 0.94 × 92). As her/his score is 47.5 (which is less than 80.81), this child would be classified as having a morphosyntactic impairment. This result is consistent with the score s/he received on the BILO (z score = −2.7 SD). On the other hand, we can consider a second (TD) child we tested (aged 6;3, that is 75 months) who obtained 97.5 on the morphosyntactic DA task. Given the cutoffs, we would consider this child as TD since her/his score is above 64.83 (=−5.67 + 0.94 × 75). This child is thus classified as TD by using our DA task, which is also consistent with her/his BILO score (z score = 0.53).

To sum up, our results revealed high sensitivity and specificity (over 80%) for our new dynamic task when considering the diagnosis of morphosyntactic impairment associated with a DLD, in particular for complex structures (clitics, passives, object relatives) as well as for the total dynamic score.

4. Discussion

Our study aimed to explore the efficacy of a new DA task of morphosyntax in diagnosing DLD in a population of French-speaking mono- and bilingual children. To recall, we assessed morphosyntactic skills by using a DA task, which provided graduated prompts to children so that they could produce the target structures. The raw scores were analyzed according to (1) linguistic group (mono- versus bilinguals) and (2) diagnostic-group (DLD versus TD). The global results confirmed that the task did not disadvantage bilinguals over monolinguals and that it distinguished TD from DLD, irrespective of the child’s linguistic group. Moreover, when looking at children with both a DLD diagnosis and a documented morphosyntactic deficit, our analyses revealed that our DA task was discriminative, as we obtained high sensitivity and specificity, especially on complex structures.

4.1. Which Variables Affect the Children’s Performances on Our Syntactic DA Task?

The results showed no effect of linguistic group (i.e., bilingualism) both for TD and DLD children on DA performance. This finding is consistent with the results of Hasson et al. (2012). More specifically, mono- and bilinguals, within their diagnostic groups, did not differ on any tested structure, even the most complex ones, such as those with accusative clitic pronouns or with object relatives. This result is encouraging and suggests that our task may be used clinically with bilingual children. DA has indeed been recommended for the assessment of bilinguals (Paradis et al. 2021). Having available a new morphosyntactic DA task would also fill a gap in diagnostic tools for this population, as recently underlined for example by Swiss SLTs (Stanford et al., forthcoming).

Secondly, our results revealed that the diagnostic group was a variable that impacted DA performance. These findings are consistent with previous studies focusing on syntax and DLD that consistently showed that DLD children performed less well than their TD peers (Delage and Frauenfelder 2020; Fleckstein et al. 2018; Paradis et al. 2003). It should be noted, however, that when performance for each individual structure was considered, the diagnostic group did not emerge as significant in explaining SVO-sentences, in which a ceiling effect was observed, with both TD and DLD children scoring very high. Such simple sentences are indeed acquired very early by children with TD and DLD (Brown 2013; Slobin and Bever 1982). The scores for complex structures, however, were influenced by the diagnostic group. Our complex structures included phrasal movement and/or embedding, two properties known to be impaired in DLD in a persistent way; (Tuller et al. 2011, 2012)8, and which are hypothetically related to deficits in working memory (Delage and Frauenfelder 2020; Zebib et al. 2020). More specifically, in our study, children with DLD scored lower than their TD peers on subject and object relative clauses (as in Delage and Frauenfelder 2020 or Frizelle and Fletcher 2015), on passive sentences (as in Garraffa et al. 2018 or Stanford and Delage 2021), and on accusative clitics (see Tuller et al. 2011). These latter difficulties are considered as clinical markers of DLD in French, in monolinguals (Hamann et al. 2003; Paradis et al. 2003; Tuller et al. 2011) but also in bilinguals (Paradis et al. 2021; Vender et al. 2018). As for sentences with past tense, it is also known that DLD children struggle with the morphological tense marker in French, both monolinguals (Frenck-Mestre et al. 2008) and bilinguals (Paradis et al. 2017).

Thirdly, we found that nonverbal reasoning influenced DA scores for each syntactic structure. This could be considered as surprising given the number of studies that regularly note the absence of links between nonverbal reasoning and syntax (see Blom and Boerma 2019; Delage and Frauenfelder 2020; Wood et al. 2021). However, we had predicted this link as our task relies on (analogical) learning, which is not the case for the static assessments described in the previous studies. It has been found that learning skills may be predicted by nonverbal reasoning (Griffiths et al. 2022). Therefore, the dynamic part of our syntactic task could be impacted by nonverbal IQ, since children with higher intellectual skills could use analogous strategies gleaned from the first item of the three structures to generate the correct target structure. Indeed, since the three items of each structure follow each other, children could apply the strategy deduced from the first item to the two following items.

4.2. Does Our Morphosyntactic DA Task Diagnose Children with DLD?

Despite the group difference between children with TD and those with DLD on all complex structures, we observed an overlap between the TD and DLD scores. This could be explained by the heterogeneity of our DLD population. As previously described, 13 children with DLD obtained a z score above −1.65 SD on the standardized test of grammar, based on normative data. This score, which is not considered as pathological, does not necessarily avoid a DLD diagnosis, as morphosyntax is not always a language domain impaired in DLD (Bishop et al. 2017). The heterogeneity of the DLD group could therefore explain the low rates reported in sensitivity and specificity, when considering the DLD sample as a whole. In our following analysis, we therefore did not include the 13 DLD children without morphosyntactic impairment in our supplementary ROC curves. In this way, when we calculated the sensitivity and specificity of the test, we found cutoffs for every structure, all leading to very good accuracy (over 80%, following the classification of (Dollaghan 2007)), except for the SVO structure that displayed global ceiling effects. As our sample included children with a large age distribution (from 5 to 11), we took age into consideration in determining our cutoffs, a variable that was not included in the cutoffs calculated by Schwob and Skoruppa (submitted) and Fleckstein et al. (2018). As a matter of fact, the age ranges in these studies were rather limited (from 5 to 7 for the first one and from 5 to 8 for the second). Since we added age into our model to determine cutoffs, our task appears to be suitable for children with a wide age-range (from 5;1 to 11;9). Moreover, adding age to our cutoffs appears essential as strong age effects have been found in the mastery of complex syntactic structures after the age of 6 (see Delage and Frauenfelder 2020). In addition to the high sensitivity and specificity of our test, we had no floor or ceiling effects for the majority of our structures, in contrast to the findings of Schwob and Skoruppa (submitted). However, LR+ and LR− as well as sensitivity and specificity scores indicate moderate success in distinguishing TD from DLD, suggesting the need to make further improvements to our task (see section on “limitations”). To sum up, our results add something innovative to the field: a DA task of morphosyntax suitable for children aged from 5 to 11 years, whereas all previous studies on the DA of syntax have focused on more restricted age ranges (e.g., (Hasson et al. 2012) on children aged from 8 to 10, (Hasson et al. 2013), on children aged from 3 to 5, and (Bain and Olswang 1995) on children aged from 2;6 to 3).

4.3. Clinical Implications

To recall, our main findings are threefold. We found: (1) no differences between mono- and bilinguals, in TD children as well as in DLD; (2) significant differences between children with and without DLD; and (3) good sensitivity (following the classification of Plante and Vance 1994) for children with both DLD and documented morphosyntactic deficiency regarding all structures (except OR). All of these results could lead to a broader, clinical use of our task in the French-speaking population. If standardized, this clinical testing tool would then be able to assist in the diagnosis of DLD, especially in the bilingual population. In addition, since multiple types of prompts were provided, such as visual priming or elicitation, the use of such an assessment could inform therapy since it indicates which form of priming is the most efficient in helping a given child produce the correct complex structure. Our DA task would then allow for more individualized treatment, as recommended in speech and language therapy approaches, notably by Martinez Perez et al. (2015). An investigation of whether the prompts lead to consistent correct production over time could indicate the validity of these prompts as useful strategies/priming. Further investigation, using longitudinal training studies, could contribute to this question.

4.4. Limitations

The first limitation of our study concerns the relatively small number of participants; this number should be increased in future studies. We acknowledge that for some subgroups, the data were limited. For example, the low number of simultaneous and sequential bilinguals did not allow us to distinguish the profiles of bilinguals according to their age of exposure to French, as discussed by Paradis et al. (2021). In future analyses, we would like to take a closer look at sequential bilinguals’ performance as compared to simultaneous bilinguals. Indeed, this topic has been discussed in the literature with variable findings. Some authors report that simultaneous bilinguals may perform similarly to monolinguals (De Houwer 1995; Meisel 2009; Paradis and Genesee 1996), whereas others find that they perform less well than their monolingual peers, for example, on accusative clitic production (Hulk 1997; Hamann and Belletti 2006). Secondly, an important limitation of this study was that the DLD diagnosis in bilingual children was only based on the diagnoses performed by French-speaking SLTs and we were not able to assess the various L1s of the bilingual children, as recommended by the American Speech Language Hearing Association (https://www.asha.org/practice-portal/professional-issues/bilingual-service-delivery/#collapse, accessed on 9 November 2022). However, as shown earlier, 15 different languages were represented in our sample, which increased the difficulty of assessing all of the children in all of their spoken languages. Thirdly, we could not include all the DLD children in our second sensitivity and specificity analyses as 13 of them did not have a documented morphosyntactic deficit. We would therefore like to expand our DLD child cohort, by including only children with documented morphosyntactic deficits. Finally, we determined the validity of our test on the basis of results from a standardized test of grammar. This is a clear limitation of our study since we essentially provided evidence for the validity of the static task. However, the strength of DA is the additional information it provides on the development of language therapy for DLD children. Hence, our task can inform clinicians on the syntactic structures that can be immediately addressed in therapy and indicate the type of cues needed to facilitate the production of these syntactic structures. As said earlier, the task needs to be improved to enable better LR+ and LR− as well as accuracy scores and ultimately better discrimination between TD and DLD. Such improvement could include augmenting the difficulty of the structures tested (adding for instance clitic sequences, i.e., accusative and dative ones, structures tested in the BILO or multiple embedding as in (Delage and Frauenfelder 2020)) or the number of items proposed (e.g., four items per structure instead of three).

5. Conclusions

The present research supports the existing literature on syntactic deficits in mono- and bilingual children with DLD, offering promising results on a DA task that does not disadvantage bilinguals over monolinguals. Indeed, our results showed that our morphosyntactic DA task did not distinguish mono- and bilingual children while being sensitive to the diagnostic group (TD vs. DLD). Moreover, we documented relatively high sensitivity and specificity in identifying morphosyntactic deficits in children diagnosed with DLD and having a documented impairment on a standardized morphosyntactic test. These results could lead to the clinical use of this task, if standardized and improved. In this way, our DA task may constitute a new tool to diagnose morphosyntactic disorders in children with diverse cultural and linguistic backgrounds. Moreover, this task could form the basis for developing a morphosyntactic treatment plan in children with DLD.