Multimodal Skills, but Not Motor Skills, Predict Narrative and Expressive Pragmatic Skills in Children With Typical Development and Neurodevelopmental Disorders

ABSTRACT

To see whether communicative‐based multimodal skills (compared to non‐communicative motor skills) predicted complex language skills, this study examined the predictive power of multimodal and motor skills on narrative and expressive pragmatic abilities across two groups. Participants were children with typical development (N = 88, M _age = 5.34, 48% female) and with neurodevelopmental disorders (N = 51, M _age = 5.01, 25% female) mostly of white ethnicity (86.3%). We evaluated children's multimodal accuracy, motor skills, core language skills, and narrative and pragmatic skills. Results revealed that, in all groups, both multimodal skills and core language significantly predicted narrative (R ² = 0.569) and pragmatic skills (R ² = 0.621), while motor skills did not. These findings highlight the relevance of multimodality in the assessment of children's complex language skills.

1. Introduction

Human communication is unique compared to other species, such that when we communicate, we communicate multimodally using both our voice to express ourselves and also our hands, face, and body (e.g., Perniss ²⁰¹⁸). In oral communication, these multiple strategies work together to help convey our meaning. For example, we might wave goodbye with our hand while saying “bye‐bye,” point to something while talking about it, or frown to show disapproval or anger. This facilitates the communication between the interlocutors and ensures that the interlocutor adequately understands the desired meaning. Because of this, multimodal skills are considered to be highly communicative skills. At the same time, multimodality partly relies on motor skills, which refer to the physical functions of body muscles and are thus devoid of communicative intent (such as building toys with hands or grabbing objects). The ability to communicate multimodally emerges from the early stages of life and continues to develop until adolescence and appears in more complex discourses (McNeill ¹⁹⁹²). Given that multimodality is naturally present in human communication, theories of human cognition that incorporate multimodality have been postulated during the past decades. One such theory, the theory of embodied cognition (Adams 2010; Foglia and Wilson ²⁰¹³) states that “the body intrinsically constraints, regulates, and shapes the nature of mental activity” (Foglia and Wilson 2013, 319) and thus aims to understand how our body impacts our cognition. In support of this theory, recent studies have demonstrated that multimodality positively predicts and impacts preschool‐ and school‐aged children's cognitive and linguistic abilities (for reviews, see Hostetter ²⁰¹¹; Goldin‐Meadow ²⁰¹⁵; Vilà‐Giménez and Prieto ²⁰²¹; see also the multimodal enrichment paradigm by Mathias and von Kriegstein ²⁰²³). The current study is therefore framed within the embodied cognition paradigm and aims to tease apart relative contributions of motor skills (which do not have communicative intent) and multimodal skills (which do have communicative intent) in relation to language skills.

Research has been conducted to investigate the relation between multimodal and language skills, both longitudinally and concurrently. Longitudinal studies have suggested that early pointing gestures during the first year of life can predict much later language skills, such as vocabulary and morphosyntax (see Kirk et al. ²⁰²² for a meta‐analysis). Another longitudinal study over 3 years showed that pointing gestures at 12 months predicted word production, sentence comprehension, and sentence production at ages 2 and 3, and grammar at age 4, while walking onset did not (Lüke et al. 2019). Additionally, the use of both referential (i.e., representing the semantic content in speech) and non‐referential gestures (i.e., not representing the semantic content in speech) in narrative discourse has been shown to predict later complex linguistic abilities, such as narrative skills (Demir et al. 2015; Vilà‐Giménez et al. ²⁰²¹). Vilà‐Giménez et al. (2021) showed that non‐referential gestures produced in parent–child interactions between 14 and 58 months were precursors of later narrative abilities at age 5. Similarly, Demir et al. (2015) reported that referential gestures produced within narration at 5 years of age predicted later narrative skills at ages 6–8 years of age.

Concurrent studies have documented similar results, such as a direct positive correlation between the rate of pointing gestures and vocabulary at 10–12 months of age (Mumford and Kita 2016). Interestingly, a recent concurrent study with older children has suggested that gesture rate (i.e., the quantity of gestures) did not predict linguistic skills, but the accuracy of gestures (i.e., the quality of gestures) did (Pronina, Grofulovic, et al. ²⁰²³). This study evaluated gesture accuracy through a multimodal imitation task with 3‐ to 4‐year‐olds. Similarly, the study by Castillo et al. (2023) indicated that the accuracy with which 3‐ to 4‐year‐old children imitated multimodal cues (i.e., gestures and prosody) positively correlated with both narrative and expressive pragmatic abilities.

The abovementioned studies reported findings on children with typical development (TD), but studies concerning children with neurodevelopmental disorders (NDD), such as autism or Developmental Language Disorder (henceforth, DLD), who have impairments in language and social communication as well as difficulties with multimodal communication (American Psychiatric Association ²⁰¹³; Bishop et al. ^{2016, 2017}), may be particularly informative in elucidating the relevant contributions of multimodal skills and general motor skills to language variation. Low rates of gesture production have been reported for children with autism, both in spontaneous speech and in more complex discourse, such as narrative (for a review, see McKern et al. ²⁰²³). As for children with DLD and with severe language difficulties, studies have documented a delay in gesture production during the first years of life (e.g., Lüke, Grimminger, et al. ²⁰¹⁷; Lüke, Ritterfeld, et al. ²⁰¹⁷) and an increase later in development, even outnumbering the gesture rates of TD children (e.g., Mainela‐Arnold et al. ²⁰¹⁴). Longitudinal evidence has suggested that early pointing gestures between 12 and 48 months were predictors of (a) later vocabulary and receptive language skills in children with autism aged between 1 and 8 years (e.g., Manwaring et al. ²⁰¹⁹; Özçalışkan et al. ²⁰¹⁶; for similar findings with a cross‐sectional study, see Ökcün‐Akçamuş et al. ²⁰¹⁹) and (b) sentence repetition and word and grammar production at ages 5 and 6 in language‐impaired populations (Lüke et al. 2020). A study evaluating body gesture imitation accuracy found that these imitation abilities at age 2 predicted socio‐communicative skills at age 4, suggesting that gesture imitation could help identify children with language difficulties (Dohmen et al. 2016). In cross‐sectional studies, expressive language skills are associated with gesture rate at 12–48 months of age in autism (Manwaring et al. 2019). Research on DLD and language‐impaired populations has also suggested that there is a tight relation between multimodality and language. The study by Guiberson (2016) showed that conventional and pointing gesture rates in semi‐structured play activities correlated with the number of different words and mean length of utterance at age 2. Studies with older children using multimodal communicative tasks, such as gesture elicitation and gesture comprehension tasks have shown that accuracy in these tasks is positively correlated with non‐verbal cognition and receptive and expressive vocabulary at ages 4 to 8 (Wray et al. 2016, 2017). Particularly, Wray et al. (2017) highlighted that it was not gesture rate, but gesture accuracy that was related to vocabulary in 6‐ to 8‐year‐old children with language impairment or typical development. Nevertheless, the causal relation of the variables from these studies is unclear.

Importantly, all of these studies report a similar tendency: multimodal skills consistently predict language skills at different developmental stages in children with and without NDD. Nonetheless, studies to date have used different tasks evaluating different multimodal and language abilities. In addition, most studies with NDD populations have mostly focused on structural language skills and generally omitted the evaluation of more complex linguistic skills, such as narrative and expressive pragmatic abilities.

The link between motor skills and language development has also been studied (see González et al. ²⁰¹⁹; Leonard and Hill ²⁰¹⁴; Wilson et al. ²⁰¹⁸, for reviews). Motor skills can be classified as either gross motor skills (involving larger muscles of the body in actions such as sitting, crawling, or walking) or fine motor skills (involving smaller muscles in actions such as object manipulation or drawing). As reported in one recent systematic review (González et al. 2019), research has suggested that there is a tight relation between motor skills and language during the early stages of development until preschool. Interestingly, it has been demonstrated that there is a tight relation between gross motor skills, such as walking onset and children's interactions with their caregivers, in the sense that the walking onset allows infants to move their hands freely while in motion, enabling them to simultaneously move and communicate (Clearfield 2011; Karasik et al. ²⁰¹¹; Walle ²⁰¹⁶). Nevertheless, the review by González et al. (2019) also revealed that this relation is less clear in school‐aged populations (e.g., Muluk et al. ²⁰¹⁴; Wang et al. ²⁰¹⁴). Significant correlations between motor skills (assessed through parents' questionnaires) and social and linguistic skills have also been reported for children with autism and DLD throughout development, from the first year of life until age 11 (e.g., LeBarton and Landa ²⁰¹⁹; Sack et al. ²⁰²²; Tseng and Hsu ²⁰²³). Nevertheless, some other studies have reported no clear relation between motor skills and language in the same population (e.g., Dziuk et al. ²⁰⁰⁷ with 10‐ to 11‐year‐olds; Leonard et al. ²⁰¹⁵ with 1‐ to 3‐year‐olds). To our knowledge, few studies have directly compared multimodal and motor skills (Castillo et al. 2023; Dziuk et al. ²⁰⁰⁷; Iverson and Braddock ²⁰¹¹; Lüke et al. ²⁰¹⁹; Ramos‐Cabo et al. ²⁰²²; Stewart et al. ²⁰²⁰) and only one performed a comparative analysis across TD and NDD populations (Wray et al. 2016). For instance, the longitudinal study by Ramos‐Cabo and colleagues suggested that while autistic children's pointing gestures measured at ages 3 to 6 were significant predictors of expressive language 1 year later, object manipulation was not (for similar findings with TD children, see Lüke et al. ²⁰¹⁹; Stewart et al. ²⁰²⁰). Iverson and Braddock (2011) showed that in preschool‐aged children with DLD, there was a negative relation between the rate of gesture production and language while controlling for motor skills. The study by Castillo and colleagues with TD preschoolers showed that multimodal imitation skills were predictors of narrative and expressive pragmatic abilities, while object‐based imitation was not. Finally, Wray et al. (2016) showed that there were no differences between TD and language‐impaired 4‐ to 9‐year‐old children in their motor skills, but that those with language impairments had worse gestural performance, suggesting that multimodal skills can be impaired regardless of motor ability. These studies highlight the importance of comparing motor and multimodal skills, as they may contribute differently to language development.

Directly comparing multimodal and motor skills is highly relevant, in part because multimodality relies on motor development. Motor capacities, like gross motor coordination or oral motor control, are needed to be able to produce hand gestures, bodily movements, or facial expressions. Motor skills do not, however, serve to express communicative intent, whereas multimodal skills such as gesture do. Because of this, comparing them might help understand the contribution of merely physical actions versus purposeful communicative actions in relation to language development. Extending investigation to children with neurodevelopmental conditions may be particularly informative because they have challenges with motor and multimodal skills; in fact, difficulties with motor skills may affect the frequency and accuracy of multimodal communication.

Traditionally, researchers have considered the contributions of motor and multimodal skills in distinct clinical populations, but a transdiagnostic approach suggests there is a value in looking across diagnostic boundaries to study traits that are common in different clinical populations and continuous with typical development (Astle et al. 2022; Donolato et al. ²⁰²³). There is substantial overlap in autism and DLD, with common difficulties in multimodal, motor, and language skills. The transdiagnostic approach allows us to examine continuous relations between language and multimodal traits across the entire sample, increasing our statistical power and the clinical implications of our findings.

Finally, most of the abovementioned studies assessed the relation between gesture (mostly pointing gestures, without looking at other multimodal skills, such as multimodal accuracy) with basic language skills (such as vocabulary or core language). However, gestures have been implicated in more complex verbal activities, such as narrative and pragmatics, in which gesture can enhance communication by providing additional information relevant to conveying the intended meaning. Narrative is defined as the ability to produce an oral discourse that contains the core components of a story linked temporally and causally (e.g., Demir et al. ²⁰¹²). While often pragmatics is understood more narrowly, such as the interpretation of non‐literal meanings (e.g., metaphors: Pouscoulous and Tomasello ²⁰²⁰; Kalandadze et al. ²⁰¹⁹; irony: Fuchs ²⁰²⁴; Wilson ²⁰¹³; scalar implicatures: Hochstein et al. ²⁰¹⁷; Mazzaggio et al. ²⁰²¹; Noveck ²⁰⁰¹), in this paper we follow the model proposed by Snow and Douglas (2017) and adopt a broader conceptualization of pragmatics that understands pragmatic competence as a “cup of competences,” that is filled with the expression and comprehension of language (language functions) in relation to the social context (social cognition functions) (see also Airenti ²⁰¹⁷; Hyter ²⁰¹⁷). In this paper, we focus on the expressive part of pragmatics (henceforth, expressive pragmatic ability), which is understood as the ability to provide pragmatically appropriate answers to a set of questions that elicit different speech acts framed within different social contexts (for more details, see Section 2).

The current study aims to evaluate how multimodal skills (particularly looking at multimodal accuracy), in comparison with motor skills, predict complex communication skills, such as narrative language and expressive pragmatic abilities, across a diverse sample of children in preschool‐ and early school‐aged children. We report descriptive statistics for each group (typical autism, DLD) for reader interest, but our analyses focus on relations between variables across the population. Multimodal accuracy will be measured through different behavioral tasks involving multimodal (gesture and prosody) imitation, gesture elicitation, and gesture‐speech comprehension. This study is designed as a confirmatory investigation, focused on hypothesis testing to validate and extend prior findings. We hypothesize that across groups, multimodal skills will be a powerful and unique predictor of narrative and expressive pragmatic competence because the three skills draw on communicative intent. In contrast, we predict the relation between motor skills and narrative/pragmatic competence will be negligible, as they lack communicative intent. We expect that children NDD will have lower scores relative to TD peers in all our target measures (cf. Wray et al. ²⁰¹⁶). Nevertheless, to the extent that gesture and speech constitute a single communicative system and go hand in hand (McNeill ¹⁹⁹²), we expect that relations between multimodal skills and language will be similar in both groups.

Overall, the current study extends the existing literature and informs the theory of embodied cognition by directly comparing communicatively‐driven multimodal skills and non‐communicative motor skills and their relations to complex linguistic abilities in young children with and without language and communication needs. This investigation will substantially advance current knowledge on the connection between bodily movement and language in child development.

2. Methods

2.1. Participants

A total of 139 children participated in this study. Participants were children with TD and NDD from Catalonia, Spain, aged between 3 and 7 (see Table 1 below for the detailed characteristics of the participants). Most of the participants were White (86.3%) while the remaining participants were Hispanic (5.1%), Black (4.3%), or Arab (4.3%).

TABLE 1.

Participants' characteristics.

	TD	NDD
		Autism	DLD
N of participants (N of females and males)	88 (43 F, 45 M)	17 (5 F, 12 M)	34 (8 F, 26 M)
N of participants with a diagnosis (N of females and males)	—	13 (5 F, 8 M)	14 (2 F, 12 M)
N of participants with reported risk (N of females and males)	—	4 (0 F, 4 M)	20 (6 F, 14 M)
Age: Mean (SD)	5.34 (0.33)	4.7 (0.55)	5.16 (0.87)
Age: Range	4.75–6.17	3.5–5.5	3.92–7.33

Participant recruitment was conducted as part of a bigger project examining children's multimodal language learning during 2021 and 2023. Typically developing children were recruited from three public schools in a city in the metropolitan area of Barcelona, Catalonia. Children with neurodevelopmental conditions were recruited through certified speech‐language therapists, psychologists, and language‐specialized teachers. To participate in the study, children in the NDD group needed to meet the following inclusion criteria: (1) have a professional report certifying a diagnosis or risk of autism or Developmental Language Disorder, (2) have a minimum vocabulary size of 50 functional words, (3) be able to systematically produce 2‐word combinations, and (4) be receiving usual intervention services. Three NDD cases were recruited from the public schools, as their teachers reported that they had severe linguistic and communication difficulties and that they were receiving weekly intervention sessions outside school.

We adopted a transdiagnostic approach (e.g., Astle et al. ²⁰²²), given that there is considerable overlap in the oral language profiles of individuals with autism and/or language disorder (DLD) (Norbury and Bishop 2003; Norbury et al. ²⁰¹⁴) and emerging evidence that oral language interventions have similar focus and effects in these two groups (see Donolato et al. ²⁰²³). Therefore, the NDD group consisted of children with autism, children at risk of being diagnosed with autism, diagnosed children with DLD, children at risk of being diagnosed with DLD, and children with severe linguistic difficulties. Participants who were at risk of being diagnosed with autism or DLD had a professional report describing severe linguistic difficulties, and at the time of testing were undergoing assessments and were in the process of receiving a diagnosis. Despite the variability in diagnosis/risk profiles in the NDD group, children's narrative and pragmatic challenges were confirmed by their usual therapist. In addition, using t‐tests with the Bonferroni adjusted p‐value of 0.006, we checked that the participants from all profiles had similar scores in all the target measures of this study (see Section 2.2). None of these measures reported significant different before or after the adjustment for multiple comparisons (see Table 2 for the statistical results, with Cohen's d as a measure of effect size).

TABLE 2.

Descriptive statistics for all measures and significant differences between children with autism and DLD.

Variable	Group	M	SD	Observed range	Possible range	Statistics
Core language	Autism	34.18	17.80	4–60	0–185	t(27.46) = 0.86, p = 0.396, d = 0.27
	DLD	29.85	14.85	4–65
Narrative retelling	Autism	2.24	1.42	0–4.75	0–6	t(33.61) = −0.18, p = 0.856, d = −0.05
	DLD	2.31	1.49	0–4.5
Expressive pragmatics	Autism	20.00	16.08	0–47	0–94	t(28.88) = −0.30, p = 0.766, d = −0.09
	DLD	21.38	14.25	0–52
Multimodal imitation	Autism	29.35	10.48	13–46	0–24	t(33.41) = 0.84, p = 0.408, d = 0.25
	DLD	26.71	10.94	8–47
Single gesture elicitation	Autism	17.29	10.81	0–34	0–37	t(32.93) = 0.64, p = 0.530, d = 0.19
	DLD	15.24	11.12	0–34
Gesture‐speech integration	Autism	3.41	2.00	0–8	0–8	t(33.10) = 0.15, p = 0.884, d = 0.04
	DLD	3.32	2.07	0–8
Visuomotor precision	Autism	178.53	63.37	0–263	0–450	t(29.72) = −0.65, p = 0.522, d = −0.20
	DLD	190.38	58.08	101–331
Manual shape imitation	Autism	9.65	4.31	4–17	0–24	t(33.44) = 0.02, p = 0.982, d = 0.01
	DLD	9.62	4.51	4–18

This study was approved by the Research Ethics Committee at the Universitat Pompeu Fabra (ref.: 228). Before the beginning of the study, legal guardians were provided with written information about the project and then gave written consent for their child to participate in the study, including video recordings of the testing sessions.

2.2. Materials

To evaluate children's linguistic, narrative, expressive, pragmatic, multimodal, and motor skills, several tasks were administered to the participants. The subsections below will describe each of the tests used.

2.2.1. Core Language Skills

Children's core language abilities were assessed using the Clinical Evaluation of Language Fundamentals (CELF); either CELF Preschool‐2 or CELF‐5 (Wiig et al. 2009, 2013) due to the differences in children's ages. CELF‐5 was administered to all TD children and NDD children older than 6, while CELF‐2 was administered to all NDD children younger than 6. For both tests, we calculated the Core Language Score involving the subtests Basic Concepts, Word Structure, Recalling Sentences and Concepts, and Following Directions from CELF‐2 and the tests Sentence Comprehension, Word Structure, Recalling Sentences, and Formulated Sentences from CELF‐5. The administration of this test lasted around 30 to 40 min per child and followed the standard CELF protocol.

2.2.2. Narrative Skills

Narrative skills were measured through a narrative retelling task, which consisted of three wordless cartoon videos and a sequence of images. First, the child watched a 40‐s cartoon featuring a single character (from the German series Die Sendung mit der Maus, https://www.wdrmaus.de/) and was asked to stand up and retell it. This procedure was repeated with another clip from the same series but this time featuring two characters and then with a 2½‐min video cartoon about a capybara (from the Colombian series Chigüiro, https://maguare.gov.co/chiguiro/). These cartoons were chosen to have a variety of stories but, importantly, despite being different cartoons, they all had the same narrative structure in which the character encountered a problem and tried to solve it until he or she was able to find a solution. After these three retellings, the examiner presented a sequence of five images featuring a child as the main character from the CUBED assessment (Petersen and Spencer 2016) and asked the child to pay attention as she would explain a story. After that, she asked the child to retell the story. The four narrative retellings were coded using a 0 to 6 coding (adapted from Demir et al. ²⁰¹⁵) considering whether children introduced the main macrostructure elements (see Table 3 for the coding). Finally, a composite score for narrative ability was calculated as the average score of the four retellings.

TABLE 3.

Scoring criteria for narrative macrostructure (adapted from Demir et al. ²⁰¹⁵).

0	The retelling does not include any descriptive sequence
1	The retelling includes one descriptive sequence (without any temporal sequence)
2	The retelling includes an action sequence (such as the main character and the problem)
3	The child produces an incomplete narrative that lacks two or more of the macrostructure elements (character, problem, attempt, solution, final)
4	The child produces an incomplete narrative that lacks one of the macrostructure elements (character, problem, attempt, solution, final)
5	The child produces a complete narrative that includes all macrostructure elements.
6	The child produces a complete narrative that includes all macrostructure elements and also adds details about the story

2.2.3. Expressive Pragmatic Ability

Children's expressive pragmatic ability was measured using the validated Audiovisual Pragmatic Test (Pronina, Prieto, et al. ²⁰²³) evaluating children's ability to appropriately produce a range of speech acts framed within different social daily life contexts. The test was a discourse completion task composed of 47 items (plus two familiarization items) that consisted of a picture representing a social situation and a short scenario explaining the situation. The situations aimed to elicit a certain speech act from the child (e.g., basic expressives –such as greetings or vocatives–, complex expressives –such as apologies–, assertions and requests). The examiner showed the picture to the child and read aloud the scenarios, and asked the child what he or she would say in that situation. Each response was given a score ranging from 0 to 2 for its pragmatic appropriateness to the intended speech act. Provided that it was a discourse completion task and that the children's responses were elicited, each item had a specific scoring criterion detailing how the response should be evaluated (refer to Table 4 for examples). Overall, a score of 2 indicated that the child provided a highly adequate pragmatic response that corresponded to the target speech act described in the scoring criterion. A score of 1 indicated that the child had expressed the adequate target speech act, but his or her response lacked pragmatic appropriateness, while a score of 0 indicated that the child did not utter any response or the response given did not correspond to the intended speech act, for example the child commented what he or she liked about the picture. Therefore, although some of the potential responses could be considered pragmatically appropriate in other contexts (e.g., spontaneous speech), they were coded as inappropriate or lacking appropriateness because they did not address the prompted sentence (e.g., the prompt explicitly asking the child to produce a question and the response being a statement). One example would be one in which the child is prompted to ask for a piece of cake and his or her response to the prompt is “I like chocolate cake” (see the last example on Table 4). Finally, an overall sum score of the 47 items was calculated. For more detailed information about the test including psychometric properties, see Pronina, Prieto, et al. (2023) or check the materials at the Open Science Framework.

TABLE 4.

Scoring criteria for expressive pragmatic ability using the Audiovisual Pragmatic Test (Pronina, Prieto, et al. ²⁰²³).

Speech act	Scoring criteria	Example	Possible responses
			0	1	2
Basic expressive (e.g., greeting)	It should be an appropriate expression of greeting and introduction to a person you just met following the structure provided	Imagine that there is a new kid in your school. You really want to meet him. When you find him, how would you say “Hello!” and your name?	Do you want to play?	Hello	Hello, my name is [NAME]/Hi, I am [NAME]. What's your name?
Basic expressive (e.g., vocative)	It should be an appropriate expression to call for a friend.	Imagine that you went to your friend Maria's house, but when you are inside you cannot see her. You think that she must be in her room. You call her	Bye‐bye!	I just arrived/I can't see you	MA‐RI‐A!/Maria, where are you?/Hello, Maria, I am here!
Complex expressive (e.g., apology)	It should be an appropriate expression to apologize to an older relatively unfamiliar adult after breaking something valuable	Imagine that you're at a friend's house and you break a nice and valuable vase. Imagine that I'm your friend's grandfather. When I come home, what would you say to me?	I didn't do it/Don't be mad at me	Sorry/I broke the vase	I am so sorry/I truly apologize/Sorry, I did it accidentally
Assertion	It should be an appropriate expression to highlight an error from an adult	Imagine that you do not like bananas but your mother gives you one. She is very sure you like them. You want to tell her that you do not like bananas. What would you say?	But I do like bananas/I want an apple	No/I don't want	Thank you, but I don't like bananas/I don't like them, is there anything else for dessert?
Request	It should be an appropriate request for cake or statement of wanting some cake	Imagine that your auntie is cutting a piece of cake and you are super hungry. How would you ask her to give you some cake?	I don't want cake/I like chocolate cake	I want cake/Give me	Can I have a piece?/Can you cut me a piece?

2.2.4. Multimodal Skills

Multimodal skills were assessed with three tests evaluating multimodal accuracy, both productively and comprehensively. First, we used a multimodal imitation task (Castillo et al. 2023) in which the child was asked to imitate a set of multimodal (gesture + prosody) cues produced within a short discursive context. The test included 12 items (plus a familiarization item) featuring referential gestures accompanying exclamations, questions, or interrogatives. First, the participant watched a short clip of an actor interacting with a puppet and was asked to look at how the examiner imitated it with a real puppet. Then he or she was asked to imitate it, also with a real puppet. Specifically, the child was asked to imitate what she did (gesture), what she said (lexicon), and how she said it (prosody). Imitation was evaluated at the gestural, lexical, and prosodic levels with a score from 0 to 2 according to the imitation accuracy. For the current study, we only used the scores from the gestural and prosodic imitation. See Castillo et al. (2023) or check the materials on Open Science Framework for more details about the task.

Second, we evaluated children's gesture accuracy using the single gesture elicitation task used by Wray et al. (2017). In this task, the child was asked to describe eight semantically‐relevant objects, actions and emotions (train, guitar, sleepwalking, sad, climbing a ladder, monkey, painting, and sword fight) without speaking, but rather using his or her body language (hands, face and limbs). Child's gestures were evaluated with a numerical score according to the accuracy of the gesture produced. One‐part gestures were coded with a maximum score of 3 (i.e., guitar, sad, painting), while two‐part gestures (i.e., involving multiple separate movements) were coded with a maximum score of 6 (i.e., train, sleepwalking, monkey and sword fight). The item climbing a ladder was coded with a maximum score of 7 in Wray et al.'s study as it can involve both arms and legs. For the current study, this item was coded with a maximum score of 4, given that Catalan and Spanish have the same word for ladder and stairs (i.e., “escala” in Catalan and “escalera” in Spanish). Therefore, this caused some children to produce gestures referring to a ladder (which involved both arms and legs), while others just represented the action of going up or down the stairs (generally only involving the legs). In order not to disadvantage those who represented stairs, the coding was adapted so that children would be awarded 4 points on that item if they accurately reproduced either the ladder or the stairs. For more details, see Wray et al. (2017).

Finally, the third measure used for multimodal skills was the gesture and speech integration subtest from PleaseApp (Andrés‐Roqueta et al. ²⁰²⁴), a digital tool evaluating different pragmatic abilities. This test evaluated the child's accuracy in comprehending gesture‐speech lexical integration. This task was presented as a game in which the participant was the main character and he or she was at a circus and needed to help a clown and a ballerina find their monkey friend who had disappeared. The child was presented with eight items (plus one familiarization item) in which the clown and the ballerina gave clues on how to find the monkey. The participant needed to guess what the clown and the ballerina were talking about by paying attention to both what they said (lexicon + prosody) and what they did with their hands (gesture). Therefore, for each item, the ballerina or clown would do something with the hand (such as moving up and down the index and middle fingers as if they were scissors) and produce a sentence (such as “I was cutting stuff for the show”). After that, the child chose from a set of three pictures which was the one that corresponded. The pictures included the accurate correct response (i.e., scissors), but also less accurate responses, such as a lexical competitor (e.g., knife) and a gestural competitor (e.g., clothespin). The responses were summed to derive a total score. This test was found to have adequate reliability (i.e., Cronbach's alpha scores ranging from 0.710 to 0.819) and validity (i.e., with significant Pearson correlations with other language and pragmatics tests ranging from 0.427 to 0.686) (for more details see Andrés‐Roqueta et al. ²⁰²⁴).

2.2.5. Motor Skills

Two subtests were administered from the sensorimotor domain from the NEuroPSYchological assessment (NEPSY‐II; Korkman et al. ²⁰⁰⁷): the visuomotor precision task and the manual shape imitation task. During the visuomotor precision task, the child was asked to grab a pen or pencil and trace a curved path, such as taking a train to a station through the rails. Each participant was instructed to do it as fast as he or she could, but to be very careful not to exceed the path borders. This task was evaluated in terms of the time it took the participant to complete the path, the number of times he or she went out of the path, and the number of times he or she lifted the pen or pencil. Second, during the manual shape imitation task, the child was asked to imitate with his or her dominant hand what the examiner was doing with her hand, paying attention to the shape and all the fingers. The participant was given up to 20 s to imitate each hand shape that the examiner produced. Whenever he or she finished with the dominant hand, the procedure was repeated for the non‐dominant hand. An overall score was calculated for each task following the standard NEPSY‐II procedure.

2.3. Procedure

Children were individually tested at their school or speech‐language therapy center. Four children from the NDD group were tested at their homes as they usually received intervention sessions at home. This ensured that children were in a familiar space and felt comfortable with the testing environment.

Testing was done in three sessions of 40 to 45 min each, with at least a week between sessions. On the first day, children were administered the CELF. Second, they were administered the narratives and the APT, and on the third day, they were administered the multimodal and motor tasks. Testing was conducted in four or five shorter (20‐ to 30‐min) sessions when children had difficulty attending for longer periods.

All tasks were presented in the form of a game to motivate the children to actively participate in the tasks. Positive reinforcement and encouragement were always provided to guarantee children were feeling motivated and confident.

Testing at the schools and data coding was conducted by the first author and five research assistants, while the testing and data coding with the children in the NDD group was conducted by the first author and a certified psychologist. Experimenters were not blind to the child group (i.e., TD or NDD).

2.4. Inter‐Rater Reliability

Inter‐rater reliability was calculated for the non‐standardized measures, which are narrative skills, expressive pragmatic ability, and multimodal skills (i.e., multimodal imitation task and single gesture elicitation task), using data from 29 participants (corresponding to 20.86% of the data). After a roughly 2.45‐h training session (1 h for expressive pragmatics, 30 min for narratives, 45 min for multimodal imitation, and 30 min for single gesture elicitation), all of the data was annotated between two coders, who were blinded to the child's diagnosis.

Inter‐rater reliability was coded using Cohen's kappa (weighted kappa), using the original scores and the scores from the coders. For the expressive pragmatic coding (N of responses = 1363), the results showed a Cohen's kappa of 0.81 (i.e., strong agreement as per McHugh ²⁰¹²). As for narratives (N of responses = 116), the agreement between the coders was strong with a Cohen's kappa of 0.86. Overall reliability for the multimodal imitation task (N of responses = 696) was also strong with a Cohen's kappa of 0.80. Regarding the single gesture elicitation coding (N of responses = 232), results of the test showed a Cohen's kappa of 0.94 (i.e., almost perfect agreement).

2.5. Statistical Analyses

To analyze the relation between, on the one hand, narrative skills and expressive pragmatic ability, and on the other, multimodal and motor skills, two sets of analyses were conducted using R (R Core Team ²⁰²¹). First, we conducted bivariate correlation analyses using Spearman's correlation coefficient, given that all variables were non‐normally distributed. Composite scores for multimodal (scores from the multimodal imitation task, single‐gesture elicitation task, and gesture‐speech integration task) and motor skills (visuomotor precision task and manual shape imitation task) were also calculated and included in the correlation analyses. To account for multiple comparisons, we applied the Benjamini‐Yekutieli (BY) correction. Two separate correlational analyses were conducted for the TD and NDD groups, respectively.

Second, to confirm our hypothesis that multimodal skills (in comparison to motor skills) could predict narrative and expressive pragmatic abilities, we used two separate multiple regression models, one with narrative as the outcome variable and another one with expressive pragmatic ability as the outcome. To simplify the models, we used composite scores for the multimodal and motor measures. Before running any regression analyses, predictor variables were standardized using the scale function in R, as they all had different scales. Also, we checked for collinearity across the predictors (κ = 2.311), indicating no multicollinearity (cf. Baayen ²⁰⁰⁸). Narrative and expressive pragmatic raw scores were introduced as the dependent variable in each model, while multimodal skills, motor skills, core language skills, group, and age were included as predictors. Also, a two‐way interaction between multimodal skills and group was included in the regression models. In the two multiple regression models, we used the adjusted R ² as the variance measure.

3. Results

The required sample size of this study was determined using G*Power Version 3.1.9.6 (Faul et al. 2009). The post hoc achieved power was calculated for the two multiple regression analyses that were performed with the inputs of 0.05 for alpha, 139 for total sample size, and 6 for the number of predictors. The level of achieved power (1 − β error probability) to detect a small effect of 0.25 was 0.997, which is above the recommended acceptable level of 0.8 (Cohen 1988).

3.1. Descriptive Statistics and Correlational Analyses

Table 5 reports the descriptive statistics for all measures as well as significant differences between groups calculated with t‐tests with the Bonferroni adjusted p‐value of 0.006 and Cohen's d as a measure of effect size.

TABLE 5.

Descriptive statistics for all measures and significant differences between the groups.

Variable	Group	M	SD	Observed range	Possible range	Statistics
Core language	TD	69.88	15.96	18–103	0–185	t(105.21) = −13.80, p < 0.001, d = −2.42
	NDD	31.29	15.85	4–65
Narrative retelling	TD	4.03	0.84	1.25–5.5	0–6	t(69.84) = −7.85, p < 0.001, d = −1.58
	NDD	2.29	1.45	0–4.75
Expressive pragmatics	TD	47.56	14.48	4–73	0–94	t(103.07) = −10.34, p < 0.001, d = −1.83
	NDD	20.92	14.74	0–52
Multimodal imitation	TD	40.74	6.31	15–48	0–48	t(70.28) = −7.97, p < 0.001, d = −1.60
	NDD	27.59	10.76	8–47
Single gesture elicitation	TD	24.33	7.08	0–37	0–37	t(74.65) = −4.92, p < 0.001, d = −0.97
	NDD	15.92	10.95	0–34
Gesture‐speech integration	TD	4.36	1.32	1–7	0–8	t(75.09) = −3.19, p = 0.002, d = −0.63
	NDD	3.35	2.03	0–8
Visuomotor precision	TD	206.61	53.97	116–373	0–450	t(96.42) = −1.99, p = 0.049, d = −0.36
	NDD	186.43	59.52	0–331
Manual shape imitation	TD	16.10	3.75	6–24	0–24	t(91.57) = −8.81, p < 0.001, d = 1.62
	NDD	9.63	4.40	4–18

Tables 6 and 7 below report bivariate correlations among the different measures for the TD group and NDD group, respectively. First, the analyses showed that composite multimodal skills positively correlated with narrative skills in both TD and NDD groups. Multimodal skills were also found to significantly positively correlate with expressive pragmatic ability in both groups. Composite motor skills did not correlate with either narrative skills or expressive pragmatic ability.

TABLE 6.

Bivariate correlations among different measures for the TD group.

		1	2	3	3.1	3.2	3.3	4	4.1	4.2	5
Core language	—	0.32*	0.44***	0.42**	0.15	0.42***	0.32*	0.04	0.02	0.14	0.13
1. Narrative retelling		—	0.42***	0.35*	0.3	0.28	0.26	−0.04	−0.05	0.2	0.08
2. Expressive pragmatics			—	0.48***	0.27	0.4**	0.09	0.08	0.06	0.25	0.09
3. Composite multimodal				—	0.64***	0.82***	0.29	0.08	0.06	0.27	0.22
3.1. Multimodal imitation					—	0.18	0.11	−0.06	−0.06	0.08	−0.1
3.2. Single gesture elicitation						—	0.14	0.13	0.11	0.23	0.42**
3.3. Gesture‐speech integration							—	0.07	0.05	0.24	0.01
4. Composite motor								—	1***	0.06	0.05
4.1. Visuomotor precision									—	−0.01	0.02
4.2. Manual shape imitation										—	0.39**
5. Age											—

^* p < 0.05.

^** p < 0.01.

^*** p < 0.001.

TABLE 7.

Bivariate correlations among different measures for the NDD group.

		1	2	3	3.1	3.2	3.3	4	4.1	4.2	5
Core language	—	0.36	0.53**	0.25	0.14	0.28	0.27	−0.22	−0.24	0.23	0.16
1. Narrative retelling		—	0.61***	0.58***	0.46**	0.55***	0.26	−0.27	−0.3	0.31	0.45*
2. Expressive pragmatics			—	0.52**	0.41*	0.43*	0.44*	−0.26	−0.3	0.41	0.46*
3. Composite multimodal				—	0.82***	0.81***	0.33	−0.25	−0.28	0.33	0.44*
3.1. Multimodal imitation					—	0.3	0.18	−0.32	−0.33	0.3	0.37
3.2. Single gesture elicitation						—	0.26	−0.09	−0.12	0.3	0.39
3.3. Gesture‐speech integration							—	−0.16	−0.17	0.14	0.1
4. Composite motor								—	0.99***	−0.12	0.1
4.1. Visuomotor precision									—	−0.2	0.06
4.2. Manual shape imitation										—	0.28
5. Age											—

^* p < 0.05.

^** p < 0.01.

^*** p < 0.001.

The correlation analyses also revealed that core language skills positively correlated with multimodal skills, but not motor skills, in the TD group. In the NDD group, no significant correlation was found between core language scores and multimodal skills or motor skills. In addition, we observed that core language positively correlated with narrative skills in the TD group, but not in the NDD group. Core language also positively correlated with expressive pragmatic ability in both groups. We also found that age positively correlated with multimodal skills in the NDD group and that age did not significantly correlate with motor skills in any of the groups.

3.2. Multiple Regression Analyses

3.2.1. Narrative Skills

The first multiple regression model predicted narrative skills from multimodal skills, motor skills, core language ability, group (NT vs. NDD), age, and the interaction between multimodal skills and group as predictors. This model accounted for 56.86% of the variance (R ² = 0.569, F(6, 132) = 31.32, p < 0.001). The model showed that multimodal skills predicted narrative performance, while motor skills did not. Core language skills and age were also significant predictors of narrative skills. Neither group nor the interaction between multimodal skills and group was found to be significant predictors (see Figure 1 for a visual representation and Table 8 for all the statistical results). This result confirmed our hypothesis that multimodal skills would be a strong predictor of narrative skills in both children with TD and NDD, as opposed to motor skills, which were not found to be significant predictors.

FIGURE 1.

The relation between multimodal skills (upper left panel), motor skills (upper right panel), language skills (lower left panel), age (lower right panel), and narrative performance, as estimated by multiple regression analysis. See Supporting Information for this same graph representing similar relations for participants with autism and DLD.

TABLE 8.

Outcome of multiple regression analysis results for narrative skills.

Predictors	β	Standard error	t	p
Multimodal skills	0.684	0.134	5.97	< 0.001
Motor skills	−0.048	0.081	−0.588	0.558
Core language skills	0.342	0.128	2.667	0.009
Group	0.424	0.266	1.594	0.113
Age	0.194	0.087	2.231	0.027
Multimodal skills × group	−0.367	0.210	−1.747	0.083

3.2.2. Expressive Pragmatic Ability

The second model assessed which variables could predict expressive pragmatic ability. As in the previous model, multimodal skills, motor skills, core language skills, group and age as well as the interaction between multimodal skills and group were included as predictors. This multiple regression model explained 62.13% of the variance in the data (R ² = 0.621, F(6, 132) = 38.74, p < 0.001). The model revealed that multimodal skills were significant predictors of expressive pragmatic ability and that motor skills were not. Results showed that core language skills also predicted pragmatic competence. However, no significant prediction was found from age, group, nor the interaction between multimodal skills and group. See Figure 2 for a visual representation of the results and Table 9 for all the results from the multiple regression analysis. This finding also confirmed our initial hypothesis about the predictive role of multimodal skills (vs. motor skills) on expressive pragmatic ability in both children with TD and NDD.

FIGURE 2.

The relation between multimodal skills (upper left panel), motor skills (upper right panel), language skills (lower left panel), age (lower right panel), and expressive pragmatic performance, as estimated by multiple regression analysis. See Supporting Information for this same graph representing similar relations for participants with autism and DLD.

TABLE 9.

Outcome of multiple regression analysis results for expressive pragmatic ability.

Predictors	β	Standard error	t	p
Multimodal skills	4.898	1.759	2.786	0.006
Motor skills	−0.630	1.067	−0.590	0.556
Core language skills	7.716	1.680	4.593	< 0.001
Group	6.362	3.481	1.828	0.070
Age	1.941	1.137	1.708	0.090
Multimodal skills × Group	2.363	2.754	0.858	0.392

4. Discussion and Conclusions

The current study is the first to jointly investigate whether both multimodal and motor skills predict narrative skills and expressive pragmatic ability in children with TD and NDD. Our findings indicate that for both populations, although motor skills are vulnerable in NDD, they do not seem to be associated with or predictive of narrative skills and expressive pragmatic ability. Instead, multimodal skills, which tap communicative intent, are unique and important predictors of children's ability to build a complex narrative discourse containing its core structural elements and to appropriately answer different pragmatic contexts. We also show that multimodal skills are predictive of narrative and expressive pragmatic competence even after core language ability is taken into account (cf. Pronina, Prieto, et al. ²⁰²³). Additionally, our findings demonstrate that diagnostic group does not predict additional variance when core language and multimodal skills have been taken into account. This indicates that there are no other “diagnosis” specific variables that improve prediction of these complex language skills, consistent with the transdiagnostic approach. These findings thus suggest that it is important to not only assess children's language skills but also their multimodal skills, as these two provide additional information that might inform clinical programs.

The results of the present study contribute to expanding research on multimodal development by providing more evidence of the robust link between multimodal skills and language (see Goldin‐Meadow ²⁰¹⁵; Hostetter ²⁰¹¹; Vilà‐Giménez and Prieto ²⁰²¹ for reviews). Interestingly, as opposed to other studies that evaluated a single multimodal ability with a single measure, the current study evaluated multimodal ability in terms of accuracy measured with different tasks, evaluating children's accuracy in tasks involving imitation, production, and comprehension of multimodal cues. Our results are in line with the findings that reveal a strong relation between multimodal accuracy and complex language skills (Pronina, Grofulovic, et al. ²⁰²³; Wray et al. ²⁰¹⁷). Importantly, we believe that it would be relevant for future studies to assess potential variations in the predictive power of different multimodal abilities.

In addition, our study also elucidates previously unclear results by demonstrating that motor skills do not predict narrative skills and expressive pragmatic ability in either group, even though children with NDD had lower scores on motor tasks relative to peers. As suggested by the recent review by González et al. (2019), this finding indicates that the relation between language and motor development is less prominent as children get older (e.g., Muluk et al. ²⁰¹⁴; Wang et al. ²⁰¹⁴) even though a stronger relation exists during the very first years of life (e.g., Clearfield ²⁰¹¹; Karasik et al. ^{2011, 2014}; Walle ²⁰¹⁶). In our study, participants' mean age was 5.22, ranging from 3 years and 6 months to 7 years and 4 months, an age range above more than half of the studies included in González et al. (2019) that targeted children younger than age 3. Our findings are similar to those reported by Wang et al. (2014), who found that motor skills at age 3 did not predict language skills at age 5, and to those that compared motor and gestural skills, indicating that gestures were predictors of later language skills, while pure motor skills like object manipulation were not (Ramos‐Cabo et al. ²⁰²²).

Second, it should be taken into account that a minimum motor ability is required to be able to produce gestures accurately, suggesting that motor and multimodal skills are not completely independent. This might be of high relevance for children with neurodevelopmental conditions, who tend to face challenges with motor skills, which might in turn affect their ability to engage in multimodal tasks. Despite the potential dependence between motor and multimodal skills, as observed in our data, motor skills and multimodal skills were not significantly correlated in either the TD group (see Table 6 above) or NDD group (see Table 7 above). This finding indicates that the tasks used did not measure the same construct and that they could be considered two distinct measures. At this point, we need to acknowledge that the tasks used in this study were selected to reflect purely motor versus multimodal communicative abilities. In contrast to other motor tasks or parental questionnaires (such as walking onset, which frees the hands, enabling children to simultaneously move and communicate—e.g., Clearfield ²⁰¹¹; Karasik et al. ²⁰¹¹; Walle ²⁰¹⁶), the motor tasks used in this study evaluated motor abilities in an exclusive way, e.g., hand precision in drawing a continuous line in a curved path and in imitating arbitrary hand configurations. Conversely, multimodal tasks were always contextualized in a communicative setting. Thus, a crucial aspect of our comparison between multimodal and motor skills is the (or lack of) communicative component they have: while motor skills refer to fine motor control and coordination without any communicative intent, multimodal skills are naturally communicative and help us express ourselves. Our findings thus suggest that the adequate expression of intentions through communicative‐based body movements (as opposed to motor control abilities) are important predictor of complex communicative language.

Although the NDD group had overall lower scores than the TD group on all measures, the grouping variable in our statistical models did not explain additional variance in the prediction of either narrative skills or expressive pragmatic ability, indicating that the underlying skills are more explanatory than the diagnostic labels. These findings are also consistent with a transdiagnostic perspective that sees neurodevelopment differences as continuous with neurotypical performance. We observed that children with NDD scored significantly below their TD peers in the multimodal tasks, a finding which is consistent with previous research (e.g., Iverson and Braddock ²⁰¹¹; Lüke et al. ²⁰²⁰; Manwaring et al. ²⁰¹⁹; Özçalışkan et al. ²⁰¹⁶; Wray et al. ²⁰¹⁶). Nevertheless, although their multimodal scores were lower, the relation with language (narrative and expressive pragmatic) measures was the same as the children with typical development. These results are consistent with reported findings on younger children, such as Özçalışkan et al. (2016) who evaluated the longitudinal relation of early pointing gestures and vocabulary 1 year later and showed similar patterns for both participants with autism and TD. We did not explicitly test the differences between children with autism and DLD, in part due to our small clinical sample which limits our statistical power (e.g., Guiberson ²⁰¹⁶; Lüke et al. ²⁰²⁰; Wray et al. ²⁰¹⁷). It may be that larger sample sizes may reveal more subtle differences in multimodal behavior and accuracy between clinical groups.

The findings of the current study contribute to the theory of embodied cognition, which posits that our body is the primary agent of our cognitive processes. The theory refers to the body and physical actions, but it remains ambiguous regarding whether body actions involve purely motor skills or whether they also encompass communicative multimodal skills. Various studies in the field following this paradigm have investigated either motor or multimodal skills, suggesting a relationship between the body and cognition. Our study expands on the embodied cognition theory by directly comparing these two sets of skills with the aim of elucidating whether the communicative nature of multimodal skills has a stronger predictive power on children's language abilities. Indeed, this study has demonstrated that this is the case: when assessing preschool‐ and school‐aged children's complex language skills, such as narrative and expressive pragmatic abilities (which are framed in a communicative rich context) it is multimodal skills, and not motor skills, that are strong predictors. This further highlights the fundamental contribution of the communicative intent in multimodal actions in this relation. Furthermore, the study revealed that the predictive relation between multimodal skills and complex language was consistent in both children with and without NDD. This transdiagnostic approach prioritizes children's strengths and needs over their diagnostic classification (Astle et al. 2022; Bishop et al. ^{2016, 2017}).

Some limitations need to be accounted for in the present study. First, our results are based on concurrent relations between the measures, as children were measured at a specific time point in their development. Therefore, we cannot show temporal precedence to assert that multimodal skills are causally related to language. In line with this, we cannot know whether the reported relation between multimodal skills and narrative and expressive pragmatic abilities would be the same if assessed longitudinally. For this, more longitudinal studies need to be conducted evaluating whether multimodal skills at earlier points in time are related to complex language competence at later points in time. Particularly, these studies should consider older children (i.e., preschool and early school), as most longitudinal studies target children at a very early age (i.e., during the first 2 years of life). Assessing older populations longitudinally would allow us to evaluate the link of more complex multimodal skills (such as multimodal accuracy rather than only gesture rate, or more gestural categories rather than only pointing gestures) with more complex linguistic skills (such as narrative discourse and expressive pragmatic ability, rather than vocabulary or core language skills). It would also allow us to test the direction of effects –our premise is that multimodal skills facilitate language acquisition, but some aspects of multimodal communication, for example gesture accuracy, rely on well‐developed semantic representations of the target. Thus, impaired language may disrupt the accuracy of some multimodal cues. Second, the motor tasks used in this study on purpose lacked communicative intent in order to differentiate them better from multimodal skills, whereas some measures used in previous studies assessed communication through motor abilities, such as directing parents' attention to objects while walking (e.g., Clearfield ²⁰¹¹). For this reason, it should be considered that our findings might have differed if alternative tasks had been used in the study. Additionally, within the NDD group, the number of boys and girls was unbalanced. Although there is no specific study showing that girls tend to perform differently than boys on multimodal tasks, it is well‐known that there is a general tendency for girls with neurodevelopmental conditions to be underdetected and underdiagnosed (Ahufinger and Aguilera 2022; Lockwood Estrin et al. ²⁰²¹). It should also be noted that children with autism were younger (M = 4.7, range = 3.5 to 5.5) compared to children with DLD (M = 5.16, range = 3.92 to 7.3). Even though there were no significant differences (p = 0.357) between the two groups, we acknowledge the heterogeneity of the sample as a limitation of the study. A final limitation of the study is that although our sample was enough to achieve the desired power to detect even small effects in multiple regression, it was not enough to run more complex models, such as measurement models or structural equation models that would allow for a more fine‐grained understanding of the relations between our variables after accounting for measurement error.

The results of the present investigation have important practical implications, both at assessment and intervention levels. First, our study highlights the importance of considering multimodal skills when evaluating children's linguistic skills. Although internationally accepted diagnostic criteria have a consensus in including gesture markers—involving a poor use of gestures in DLD (e.g., Bishop et al. ^{2016, 2017}) or deficits in producing or comprehending gestures in autism (American Psychiatric Association ²⁰¹³)—these guidelines do not provide specific protocols to measure multimodal abilities. This study provides insight into the evaluation of multimodal skills beyond assessing only gesture rate or gesture comprehension. For this, practitioners should consider a wider perspective including different multimodal skills (such as gesture rate, multimodal imitation accuracy, gesture comprehension, or gesture‐speech integration) framed within a linguistically relevant framework, as they can serve as early signs of language and communication disorders. Our results also have practical implications for intervention. In fact, there is growing evidence that multimodal enrichment can help language learning (Mathias and von Kriegstein ²⁰²³), in both typically developing children (e.g., Igualada et al. ²⁰¹⁷; Kartalkanat and Göksun ²⁰²⁰; Pronina et al. ²⁰²¹; see Goldin‐Meadow ²⁰¹⁵; Vilà‐Giménez and Prieto ²⁰²¹ for reviews) and children with neurodevelopmental disorders (e.g., Dargue et al. ²⁰²¹; Lüke and Ritterfeld ²⁰¹⁴; Frey and Lüke ²⁰²³). Given that we find, (a) the link between multimodal skills and linguistic competence is the same in both groups and (b) similar positive effects of multimodal training have been documented for both groups, it seems possible that similar instruction and training could be beneficial to all children with narrative and expressive pragmatic language challenges, regardless of diagnostic group. Future research that incorporates multimodality into intervention trials will enable us to answer these questions of causality and impact.

All in all, the current study supports the hypothesis that it is the intentional communicative component of body movements (i.e., multimodality) that associates with complex linguistic abilities, such as narrative and expressive pragmatic abilities, rather than the physical attributes themselves. This relation appears to be similar for both children with typical development and those with neurodevelopmental disorders. Hence, more attention should be paid to these multimodal skills throughout children's language development, as they can serve as indicators of language difficulties and may serve as educational and interactional tools to boost language abilities.

Ethics Statement

This study has been approved by the Institutional Committee for Ethical Review of Projects (CIREP) at Universitat Pompeu Fabra (ref.: 228). Written informed consent to participate in this study was provided by the legal guardians of the participants.

Supporting information

Figure S1. The relation between multimodal skills (upper left panel), motor skills (upper right panel), language skills (lower left panel), age (lower right panel), and narrative performance, as estimated by multiple regression analysis.

Figure S2. The relation between multimodal skills (upper left panel), motor skills (upper right panel), language skills (lower left panel), age (lower right panel), and pragmatic performance, as estimated by multiple regression analysis.

Data Availability Statement

The data and the analytic code necessary to reproduce the analyses presented here are publicly accessible in the Open Science Framework repository (https://osf.io/z9nku/). The materials necessary to attempt to replicate the findings presented here are not publicly accessible, but are available from the first author. The analyses presented here were not preregistered.