Contributions of early motor deficits in predicting language outcomes among preschoolers with developmental language disorder

Abstract

Purpose: We assessed the extent to which language, speech, and fine/gross motor skills in preschoolers with developmental language disorder (DLD; also referred to as specific language impairment) predicted language outcome two years later.

Method: Participants with DLD (n = 15) and typical development (TD; n = 14) completed language, speech, and fine/gross motor assessments annually, beginning as 4- to 5-year-olds (Year 1 timepoint) and continuing through 6 to 7 years of age (Year 3 timepoint). We performed Pearson correlation and hierarchical regression analyses to examine the relative contributions of Year 1 language, speech, and motor skills to Year 3 language outcome in each group.

Result: Among children with DLD, Year 1 fine/gross motor scores positively correlated with Year 3 language scores, uniquely explaining 40% of the variance in language outcomes. Neither Year 1 language, speech-sound, nor speech-motor scores predicted language outcome in this group. Among children with TD, only Year 1 language predicted language outcome.

Conclusion: This small longitudinal study reveals that, among preschoolers with DLD, certain early fine/gross motor deficits predict persistent language impairment. Future research that includes larger sample sizes and motor tasks that incorporate complex sequencing will enhance the understanding of the relationship between language, speech, and motor skills; specifically, whether certain motor deficits simply co-occur with language deficits or whether they are tied to DLD through shared impairments in sequential learning mechanisms.

Keywords:

• developmental language disorder
• specific language impairment
• motor development
• procedural learning
• speech-language pathology

Introduction

Developmental language disorder (DLD; also referred to as specific language impairment—SLI¹) affects about 7% of 5-year-old children (Tomblin et al., 1997) and negatively impacts social, academic, and occupational outcomes. Relative to their typically developing (TD) peers, school-aged children with DLD are three times more likely to be bullied and depressed, and six times more likely to experience anxiety (Conti-Ramsden & Botting, 2004). Young adults with language impairment are at significantly higher risk for learning disabilities than their TD peers (Young et al., 2002), and adults with language impairment are two times more likely than other adults to face prolonged unemployment (Law, Rush, Schoon, & Parsons, 2009). Despite the prevalence and long-term negative consequences, young children with DLD remain underserved and under-studied (e.g. Bishop, 2010; McGregor, 2020). Early identification of children at risk poses a significant clinical and empirical challenge (e.g. Pawlowska, 2014). An important strategy for earlier and more precise identification may be to look outside of the language domain. In the present study, we assessed the extent to which adding measures of speech and fine/gross motor skill to language measures improves prediction of language outcome.

Theoretical perspectives of language impairment

Traditionally, SLI is defined based on exclusionary criteria—that is, as a significant deficit in language, not explained by cognitive, perceptual, motor, socioemotional, or experiential deficits (Leonard, 2014; Rice, Wexler, & Cleave, 1995). Research and clinical efforts have emphasised the overt morphosyntactic deficits characteristic of children with DLD, and accordingly, DLD is often identified in the late pre-school years based on language assessments related to morphosyntax (e.g. Leonard, 2014). Importantly, however, norm-referenced language tests and language sample analyses are poor predictors of later language outcomes in individual children, limiting the accurate diagnosis of DLD among preschoolers (e.g. Rudolph, Dollaghan, & Crotteau, 2019). Furthermore, children with DLD exhibit weaknesses in areas beyond morphosyntax, such as speech-sound accuracy and organisation (e.g. Alt, Plante, & Creusere, 2004; Benham, Goffman, & Schweickert, 2018) and fine/gross motor skill (e.g. Hill, 2001; Hsu & Bishop, 2014; Vuolo, Goffman, & Zelaznik, 2017).

We designed the present study within a broader perspective of DLD based on alternative theoretical accounts in which the disorder is tied not only to linguistic deficits, but to more general cognitive processing deficits affecting language, speech, and motor domains (e.g. Kail, 1994; Tallal, Miller, & Fitch, 1993; Ullman & Pierpont, 2005; Vuolo et al., 2017). In these accounts, deficits observed in DLD, including those of morphosyntax, “are not specific to an abstract linguistic rule system but rather are secondary to perceptual and/or cognitive processing deficits” (Tomblin, Mainela-Arnold, & Zhang, 2007, p. 270), and “specific means idiopathic (i.e. of unknown origin) rather than implying there are no other problems beyond language” (Bishop, 2014, p. 381).

One prominent account, the Procedural Deficit Hypothesis (PDH; Ullman & Pierpont, 2005; Ullman, Earle, Walenski, & Janacsek, 2020), posits that an impaired procedural learning system underlies both the morphosyntactic and the non-linguistic deficits exhibited by individuals with DLD. The procedural system relies on several brain structures, particularly those of the corticostriatal loop, and is responsible for cognitive and sensorimotor rule- and sequence-learning. By contrast, the declarative memory system stores “idiosyncratic mappings in a memorised ‘mental lexicon’” (Ullman & Pierpont, 2005, p. 403) and is subserved by the hippocampus and other medial temporal lobe regions that connect to temporal and parietal neocortical regions. Indeed, group differences in brain structure and function have been identified—including in areas associated with procedural learning—between children with TD and DLD, though some studies reveal greater activation while others reveal less activation in children with DLD (reviewed in Mayes, Reilly, & Morgan, 2015). The PDH asserts that, for individuals with DLD, learning aspects of language that involve hierarchical combinations experienced over multiple exposures, such as certain grammatical and syntactic forms, is relatively more impaired than learning that relies on the declarative system, for example individual lexical items learned by explicit teaching. Notably, though morphosyntax and word form learning are often more affected, semantic aspects of word learning are not entirely spared in DLD (e.g. Kan & Windsor, 2010).

We contend that only some aspects of the procedural system—those related to sequential pattern learning—are affected in children with DLD (e.g. Goffman, 1999; Goffman & Gerken, 2020; Hsu & Bishop, 2014; Vuolo et al., 2017). Just as rule-governed, hierarchical aspects of language that involve procedural learning (phonology, morphology, syntax) prove difficult for children with DLD, so too do nonspeech motor tasks, particularly those which involve hierarchical and sequential demands or complex movement sequences (Tomblin et al., 2007). In what follows, we describe evidence that implicates each domain of deficit in DLD and suggest that sequential pattern learning deficits may be one factor underlying DLD.

Domains of deficit in DLD

Morphosyntax

As described above, the hallmark deficit in classic accounts of language impairment is morphosyntax (e.g. Leonard, 2014; Rice et al., 1995; Van der Lely, 1998). Undisputedly, acquisition and use of grammatical components of language, particularly of inflexional morphology required for tense and agreement, is difficult for children with DLD. Rice and colleagues (1995) reported that 5-year-old children with SLI used fewer finite verb forms than same-aged and younger TD peers, supporting the Extended Optional Infinitive account of language impairment. Similarly, Van der Lely (1998) noted the inconsistent use of inflexional forms even in a 14-year-old child with SLI, also characterising his language profile as containing “optionality in his grammar” (p. 166). While morphosyntactic deficits are a critical and central component of the DLD profile, the following sections outline additional areas that may be important components of the diagnostic category, as well.

Speech

Speech-sound production and organisation

Nonword repetition deficits, though not specific to DLD, appear to be a component of the disorder. Children with DLD perform poorly on nonword repetition tasks, especially those that incorporate sequences of increasing length and complexity (e.g. Dollaghan & Campbell, 1998; Graf Estes, Evans, & Else-Quest, 2007; but see Pawlowska, 2014). Some young children with DLD also score below the typical range on standardised assessments of articulation and phonology (e.g. Alt et al., 2004; Benham et al., 2018).

Children with TD and DLD demonstrate interactivity across speech and language systems, with lexical factors influencing speech production accuracy. Both groups produce novel words with high phonotactic frequency more accurately than those with low phonotactic frequency (e.g. Munson, Kurtz, & Windsor, 2005). Interestingly, in production, children with DLD may be more influenced by phonotactic probability than their peers with TD—percent phonemes correct (Munson et al., 2005)—and perception—lexical decision tasks (Quémart & Maillart, 2016). Furthermore, in their production of novel word and nonword forms, children with DLD are more variable and less accurate at the segmental and prosodic levels than their TD peers (e.g. Benham et al., 2018; Goffman, 1999). Benham and colleagues used network science analyses to demonstrate that children with DLD organise syllable sequences poorly relative to their TD peers, evidenced by a higher number of nodes and edges in their networks. In short, children with DLD have difficulty at multiple levels of speech-sound production and organisation.

Speech-motor variability

In addition to behavioural measures of speech-sound accuracy using transcription and perceptual approaches, children’s speech-motor organisation and variability have been studied using articulatory kinematic analyses. These analyses, which assess spatiotemporal variability of lip and jaw movement across multiple productions of a linguistic target (Smith, Goffman, Zelaznik, Ying, & McGillem, 1995), have revealed that speech-motor organisation is implicated in children with DLD (e.g. Goffman, 1999; Saletta, Goffman, Ward, & Oleson, 2018). Compared to TD peers, children with DLD demonstrate increased articulatory variability when producing novel words (e.g. Goffman, 1999). Importantly, differences between children with TD and DLD in speech-motor stability are greater when the task obligates higher linguistic load—active retrieval—compared with low linguistic load—rote repetition (e.g. Saletta et al., 2018). That is, children with DLD do not have a generalised speech motor deficit that affects all speech production tasks. Only some tasks, such as those containing high linguistic load and sequential processing, are affected.

Together, the above speech-sound and speech-motor profile suggests that aspects of phonological organisation and speech-motor skill may influence early speech production difficulties in children with DLD, consistent with our contention that complex sequential pattern deficits may be one factor underlying DLD.

Fine/gross motor skill

Children with DLD often exhibit deficits in fine/gross motor skill and praxis ability. As a group, they score lower than TD peers on standardised assessments of manual dexterity and balance (e.g. Hill, 2001; Sanjeevan & Mainela-Arnold, 2019; Zelaznik & Goffman, 2010). They also perform more poorly than their TD peers on experimental tasks requiring imitation of representational and non-symbolic gestures (e.g. Hill, 1998; Hill, 2001).

Additionally, children with DLD have more difficulty organising sequentially patterned elements, evidenced by their weaker performance on nonverbal serial reaction time tasks (e.g. Hsu & Bishop, 2014; Lum, Conti-Ramsden, Morgan, & Ullman, 2014; Tomblin et al., 2007) and bimanual coordinated and timed clapping tasks (Vuolo et al., 2017). Again, this difference in performance between TD children and children with DLD is most often observed when sequential complexity of tasks is high (e.g. Hill, 2001; Sanjeevan & Mainela-Arnold, 2019; Vuolo et al., 2017). Importantly, there are some procedural tasks that are not affected in children with DLD, such as metronomic timing (Vuolo et al., 2017; Zelaznik & Goffman, 2010) and pursuit rotor (Hsu & Bishop, 2014). On these procedural tasks, children with DLD perform similarly to age-matched peers. Therefore, it seems there is a “unique contribution of sequence to poor motor procedural learning in [DLD]” (Hsu & Bishop, 2014 p. 361). Children with DLD show weaknesses in procedural tasks that are inherently sequential (e.g. Hebbian verbal sequence learning, coordinated bimanual timing, and standard serial reaction time; SRT) and not on those procedural tasks that are non-sequential (e.g. pursuit rotor, metronomic timing). Critically, sequential patterning (such as is required for SRT tasks and rhythmic clapping) undergirds morphosyntax and phonology— domains that prove particularly difficult for children with DLD. In this way, it is possible that the language, speech-sound, and nonspeech motor weaknesses characteristic of children with DLD may all originate in and mechanistically relate via sequential pattern learning processes.

Some researchers argue that motor deficits observed in DLD are not mechanistically tied to the language deficit, but rather reflect a neuromaturational delay affecting both language and motor systems (e.g. Hill, 2001; Locke, 1994). Deficits in speed of information processing may also influence both systems (e.g. Leonard et al., 2007). However, many well-developed hypotheses, including those relying on neural underpinnings, suggest a mechanistic relationship between language and action (e.g. Arbib et al., 2014; Casado et al., 2018; Hill, 2001).

In sum, motor deficits have been identified in children with DLD. Some accounts of DLD include extralinguistic deficits that may be related to procedural memory, and perhaps specifically to the complex sequential learning that the procedural memory system appears to subserve (Goffman, 1999; Hsu & Bishop, 2014; Vuolo et al., 2017). By these accounts, incorporating measures of speech-sound production, speech-motor, and fine/gross motor skill may result in more accurate predictions of later language status than would relying on morphosyntactic measures alone. If these domains are not mechanistically tied to the disorder, they should not contribute specifically to predictions about outcome. Alternatively, if complex sequential pattern learning is a mechanism underlying the disorder, then the addition of these broad speech and motor measures should enhance predictions of outcome.

Research question

Data from a two-year (three timepoint) longitudinal study were analysed to investigate the extent to which (a) language deficits identified in 4- to 5-year-olds with DLD persist into the early school years, and (b) early measures of speech-sound accuracy/organisation, articulatory variability, and fine/gross motor skills enhance prediction of language outcomes. We hypothesised that a combination of early deficits conceptually tied to sequential patterning aspects of procedural learning would better predict language outcome than early grammatical deficits alone. That is, we anticipated that preschool-aged children with DLD who exhibited a combination of impaired morphosyntax, speech, and motor skills would be more likely to have impaired language two years later than children with DLD who exhibited equally severe early morphosyntactic deficits but not speech and/or fine/gross motor weaknesses.

Method

General design of larger longitudinal study

The present data were obtained from a longitudinal study of children with DLD and TD. Children were aged 4 to 5 years at the onset and 6 to 7 years at completion. The study, designed to investigate the relationship between language and motor development in children with DLD, was conducted while the last author was at Purdue University. Approval from Purdue University’s Institutional Review Board (IRB), parental informed consent, and child assent were obtained prior to participation. Participants from the greater community of Lafayette, Indiana, were recruited through advertisements and referrals. Approval from the University of Texas at Dallas IRB was obtained for the data analysis phases included in the present study.

Children were categorised into two groups at the onset of the study: DLD and TD. Assignment to the DLD group was determined based on standard exclusionary criteria for SLI outlined by Leonard (2014). In the language domain, this included use of the norm-referenced, standardised assessments Structured Photographic Expressive Language Test-Pre-school 2 (SPELT-P2, Dawson, Eyer, & Fonkalsrud, 2005) or The Structured Photographic Expressive Language Test: Third Edition (SPELT-3; Dawson, Stout, & Eyer, 2003). Children with DLD participated in a summer clinical and research program (co-led by Lisa Goffman and Laurence Leonard), in which the SPELT-P2 was used as the diagnostic criterion for entry. Thus, all children in the DLD group completed the SPELT-P2. The children with TD participated in the laboratory study only, not in the summer clinical program; these children were administered the SPELT-3. A standard score (SS) of ≤87 on the SPELT-P2 was used for assignment to the DLD group, a cut-off point with high sensitivity (96%) and specificity (95%) for identifying children with DLD (Greenslade, Plante, & Vance, 2009). On the SPELT-3—used to rule out a diagnosis of DLD among the TD group—a cut-off of 85 (≤1 standard deviation [SD] below the mean) was used, which has been reported to be 71.9% sensitive and 100% specific for identifying DLD (Perona, Plante, & Vance, 2005). All children in the TD group also obtained typical finite verb morphology composite (FVMC) scores. In addition, all exclusionary criteria (Leonard, 2014) were met—including normal hearing, normal nonverbal intelligence (i.e. SS of ≥ 85 on the Columbia Mental Maturity Scale; Burgemeister, Blum, & Lorge, 1972), and no history of neurological impairment (based on caregiver report) or autism (i.e. minimal to no symptoms on the Childhood Autism Rating Scale-Second Edition; Schopler, Van Bourgondien, Wellman, & Love, 2010). The structural portion of the Robbins and Klee (1987) oral mechanism examination (OME) was conducted to rule out structural deficits; all participants scored within the typical range and there were no group differences. All participants were monolingual English speakers.

Participants entered the study as 4- to 5-year-olds (Year 1 timepoint) and were seen annually at two more timepoints, completing the study in the early school years, at 6 to 7 years of age (Year 3 timepoint). At each timepoint the participants completed standardised assessments of language, speech, and fine and gross motor skill, in addition to a variety of experimental paradigms focussed on language, speech, and manual production. The specific subset of participants and assessments included in the current analyses are described below.

Current study

Participants

The current study includes 15 children with DLD (mean age at Year 1 = 59 months, SD = 5.97 months; eight males) and 14 children with TD (mean age at Year 1 = 59.57 months, SD = 6.50 months; seven males). This longitudinal sample is the subset of the Year 1 participants (DLD = 40; TD = 21) who returned at Year 3 and completed the Year 3 language outcome measure. Independent samples T-tests confirmed that Year 1 behavioural assessment scores for this longitudinal subset do not differ from those who did not return for longitudinal follow-up or did not complete the language outcome measure, for either the DLD or TD groups. That is to say, severity did not dictate which families returned for follow-up. These behavioural profiles and corresponding statistical values are listed in Table S1 of the Supplemental Materials.

Of note, in the current study, 12 of the 15 participants with DLD and all 14 of the participants with TD had scores for each of the Year 1 behavioural predictor variables. Fourteen of the 15 children with DLD and 11 of the 14 children with TD had scores for the experimental predictor variable (spatiotemporal variability of speech production movement).

Reasons for missing data include inability or unwillingness to participate in a particular task within a session and/or failure to attend a session altogether. In most cases in the follow-up years, families who dropped out simply could not be located. The number of participants included in each analysis is noted in the description of its results and/or the corresponding tables. Figure S1 of the Supplemental Materials provides a schematic of participant inclusion.

Most, if not all, of the participants with DLD were receiving or had received speech-language therapy previously, as they were recruited through school and clinical programs. While actively enrolled in the summer clinical-research program, participants were not receiving outside therapy. The intervention focus in the summer program was not related to grammatical, motor, or sequential learning; rather, to avoid overlapping with the research questions, goals targeted vocabulary development and literacy skills. Other than the summer clinical-research program, the frequency, duration, and/or focus of intervention is not known.

Instruments

Language outcome

Year 3 language ability was measured using the Core Language Score (CLS) SS, a global language measure from the Clinical Evaluation of Language Fundamentals – Fourth Edition (CELF-4; Semel, Wiig, & Secord, 2003). A SS cut-off of ≤85 defined impaired performance, which is 100% sensitive and 82% specific for identifying language impairment (Clinical Evaluation of Language Fundamentals – Fourth Edition, 2008). The CELF-4 CLS is comprised of four subtests: Word Structure, Concepts and Following Directions, Formulated Sentences, and Recalling Sentences. At Year 3, we could not use the SPELT-P2, as it is only normed through age 5;11. We instead used the CELF-4, which has better diagnostic accuracy for this age group than the SPELT-3, and is one of the most widely used clinical measures (e.g. Alt, Gray, Hogan, Schlesinger, & Cowan, 2019; Storkel et al., 2019).

Language predictors

Year 1 language ability was measured by (1) the SPELT-P2 SS (DLD group) and the SPELT-3 SS (TD group), and (2) the FVMC score obtained from 50-100 spontaneous utterances produced during a 15-minute play-based language sample. The FVMC score provides the percentage of finite verb morphemes (i.e. regular past -ed, present third person singular -s, copula and auxiliary forms of is, are, and am) correctly used in obligatory contexts, and has been found to be a marker of DLD (e.g. Leonard, Miller, & Gerber, 1999; but see Rudolph et al., 2019). A cut-off of ≤85% defined impaired performance (Gladfelter & Leonard, 2013).

Speech-sound accuracy predictors

Measures of Year 1 speech-sound production accuracy consisted of (1) the Consonant Inventory SS of the Bankson-Bernthal Test of Phonology (BBTOP; Bankson & Bernthal, 1990), and (2) total percent phonemes correct on a nonword repetition task (NRT; Dollaghan & Campbell, 1998). The BBTOP indexes the accuracy of consonant productions in single words. A SS cut-off of ≤85 was used to define impairment. In the NRT, children imitate nonsense words of increasing length from three to nine phonemes (one to four syllables). We did not classify impaired performance on the NRT given the lack of an accepted cut-off point (e.g. Dollaghan & Campbell, 1998; Paradise et al., 2005).

Fine/gross motor predictors

To assess Year 1 fine/gross motor skill, standard scores from the 3- to 6-year-old age band of the Movement Assessment Battery for Children, Second Edition (MABC-2; Henderson, Sugden, & Barnett, 2007) were used. The MABC-2, comprised of three subtests (Aiming and Catching; Balance; Manual Dexterity), is used to assess motor skill in both TD and impaired populations, and is often used as part of a battery to diagnose developmental coordination disorder—DCD (e.g. Harris, Mickelson, & Zwicker, 2015). In the current study, the MABC-2 was used as a measure of fine/gross motor ability, not to diagnose DCD.

In the Aiming and Catching subtest, children toss a beanbag onto a mat, and also catch a beanbag thrown by the examiner. The number of successful throws and catches are each recorded as raw scores that are combined to form the Aiming and Catching SS.

In the Balance subtest, children complete three tasks: balancing, toe-walking, and jumping. In the balancing task, children stand on each leg separately. The raw score reflects the number of seconds children balanced in accordance with the criteria (e.g. lifted leg must not be anchored). In the toe-walking task, children walk along a taped line with their heels raised; the highest number of consecutive steps taken without erring (e.g. dropping heels to ground, stepping off of line) is counted as their raw score. In the jumping task, children jump with their feet together along five consecutively placed mats, and the highest number of successful jumps is their raw score. The raw scores from each of these tasks are combined to form the Balance SS.

Lastly, in the Manual Dexterity subtest, children also complete three tasks: posting coins, threading beads, and trail drawing. In the posting coins task, children place plastic coins one at a time into a slotted box. The amount of time taken to complete the task is their raw score. When threading beads, children place beads one at a time onto a string; the amount of time taken to string all beads is their raw score. For trail drawing, children draw a line along a trail without lifting the pen or straying outside the boundaries. The number of errors made (e.g. lifting pen, marks outside the borders, overlapping lines, discontinuous lines) is their raw score. The raw scores for these three tasks are combined to form the Manual Dexterity SS.

In addition to the individual standard scores from each of these subtests, a composite score of all three subtests is calculated and is reflected in a Total Score. Each of the three subtests, as well as the Total Score, has a mean score of 10 and SD of 3. A score ≤1 SD below the mean indicated impaired performance (i.e. a score of 7 or below).

Speech-motor variability predictor: spatiotemporal index (STI)

To more directly assess our theoretical account of DLD as a complex sequential pattern deficit, we included an experimental task of articulatory sequence variability in the current analysis. This task had two conditions, only one of which (repetition) is included. In the repetition task, participants repeated the sentence “Mom pats the puppy” 10 times. This task was completed at three different sessions during Year 1, with each session separated by at least 24 hours. To minimise learning effects, we analysed only the first session in which the participant successfully completed the task. We considered successful completion a session in which there were at least five fluent and segmentally consistent sentence repetitions.

We analysed articulatory stability of the repeated productions as reflected in the spatiotemporal index (STI; Smith et al., 1995). Measurements of lip and jaw movement were obtained using the Optotrak Cedrus motion capture system (Northern Digital) during the sentence repetition task. Displacement trajectories for each sentence production (beginning at lip closure of the first “m” in “mom” and ending at that of the final “p” in “puppy”) were extracted from the lower lip signal based on velocity criteria. Lip aperture was determined by subtracting, point by point, the difference between the upper lip and the lower lip signal. This aperture signal was then amplitude- and time-normalised. The SD was computed successively at 2% intervals across all normalised productions and then summed. The sum of the SDs reflects the STI; larger STI values indicate more variability across sentence productions, and smaller STI values represent more stable productions. Spatial and temporal normalisation allows for the underlying motor pattern to emerge, independent of differences in rate or loudness. Thus, the STI serves as an index of articulatory sequence patterning (Smith et al., 1995).

Analyses

Descriptive statistics

To visualise the behavioural profile of the study sample, we calculated the means and SDs on each measure for the TD and DLD groups, and determined the percentage of participants in each group who scored below the normal range, defined as ≤1 SD below the mean (excepting the SPELT-P2, as described above), on each of the assessments.

Language, speech-sound, and fine/gross motor predictors of language outcome

We used Pearson correlation and multiple regression analyses to investigate the extent to which measures of language, speech-sound accuracy, and fine/gross motor skills at Year 1 (age 4 to 5 years) predicted Year 3 (age 6 to 7 years) language outcome in children identified with DLD and in children with TD at Year 1. Of specific interest was the relative contribution of Year 1 language, speech-sound, and fine/gross motor measures in predicting language outcome (i.e. Year 3 CELF-4 CLS SS). As such, hierarchical regressions (DLD n = 12; TD n = 14) were conducted to determine the change in R², with language measures comprising Block 1 (i.e. Year 1 SPELT and Year 1 FVMC); speech-sound accuracy measures in Block 2 (i.e. Year 1 BBTOP and Year 1 NRT); and fine/gross motor measures in the final block (i.e. Year 1 MABC-2 Total).

Individual developmental trajectories

Based on the results of the above analyses, we examined individual developmental profiles of participants with DLD in an effort to understand the predictive value of early fine/gross motor skills at a more fine-grained level and to provide perspective on the potential clinical relevance of the group findings.

Articulatory variability as a predictor of language outcome

In a final analysis we conducted Pearson correlation and bivariate regression analyses to more directly assess the relationship between the ability to produce stable articulatory movement sequences at Year 1 and language outcome at Year 3. Year 1 STI values from the experimental sentence repetition task were used to predict language outcome for the TD group (n = 11) and the DLD group (n = 14).

Result

Descriptive statistics

Table I contains descriptive statistics for language, speech-sound, and motor measures by group and by time point and the percentage of children who performed below the typical range on each assessment. Generally, across all measures, the children with DLD performed more poorly than the children with TD. Notably, the majority (80%) of the children with DLD performed below the typical range on the BBTOP at Year 1, but fine/gross motor performance was more variable in this group, with about 42% scoring below the typical range on the MABC-2 Total at Year 1. Longitudinally, although by definition all of the children with DLD had impaired scores on the language measure (SPELT-P2) at Year 1, only 40% scored below the typical range on the language outcome measure (CELF-4 CLS) at Year 3. By contrast, all of the children with TD scored within or above typical range on the language tests at both time points.

Table I. Mean performance on Year 1 and Year 3 language, speech-sound, & motor measures, and percentage of children with scores below normal range, by Year 1 diagnostic group (DLD, TD).

Language & speech-sound	Year 1	Year 1	Year 3	Year 1	Year 3	Year 1	Year 3
	SPELT	FVMC	CELF-4 CLS	BBTOP	BBTOP	NRT	NRT
DLD group
Mean (SD):	74.93 (9.91)	50.07 (20.84)	90.73 (14.08)	70.93 (10.10)	86.18 (7.63)	56.53 (8.39)	74.80 (7.42)
% Impaired:	100% (15/15)	93.33% (14/15)	40% (6/15)	80% (12/15)	54.55% (6/11)	n = 15	n = 10
TD group
Mean (SD):	109.07 (11.63)	91.29 (11.49)	108.79 (12.46)	99 (9.95)	N/A	72.21 (12.62)	84.31 (7.59)
% Impaired:	0% (0/14)	14.29% (2/14)	0% (0/14)	0% (0/14)	28.57% (4/14)	n = 14	n = 13
Fine/gross motor (MABC)	Year 1:	Year 3:	Year 1:	Year 3:	Year 1:	Year 3:	Year 1:	Year 3:
	Total	Total	Manual Dex	Manual Dex	Balance	Balance	Aim/Catch	Aim/Catch
DLD group
Mean (SD):	8.83 (3.97)	8.93 (3.43)	8.17 (3.56)	8.80 (3.14)	9.25 (3.82)	9.13 (3.16)	10.75 (2.93)	9.33 (4.15)
% Impaired:	41.67% (5/12)	26.67% (4/15)	41.67% (5/12)	26.67% (4/15)	33.33% (4/12)	26.67% (4/15)	16.67% (2/12)	33.33% (5/15)
TD group
Mean (SD):	11.00 (2.69)	10.43 (2.31)	9.86 (2.54)	10.64 (1.98)	12.36 (2.82)	11.29 (2.43)	10.43 (2.79)	8.93 (3.22)
% Impaired:	7.14% (1/14)	7.14% (1/14)	21.43% (3/14)	0% (0/14)	0% (0/14)	7.14% (1/14)	7.14% (1/14)	21.43% (3/14)

Notes. All Year 1 scores are predictors. Year 3 CELF-4 is the outcome measure. Additional Year 3 scores are presented here to provide a broader picture of the behavioural profiles. At Year 3, TD children completed the BBTOP only in the presence of clinical concern (n = 6, M = 87.17, SD = 6.18). The remaining TD participants (n = 8) were considered to have speech within normal limits based on clinical judgement. For NRT, percentage of impaired scores are not presented given the lack of an accepted cut-off point.

Language, speech-sound, and fine/gross motor predictors of language outcome

In the DLD group, neither Year 1 language measures (SPELT-P2 r = 0.169, p = 0.546; FVMC r = −0.239, p = 0.390), speech-sound measures (BBTOP r = −0.270, p = 0.330; NRT r = 0.169, p = 0.546), nor the Aiming and Catching subtest of the MABC-2 (r = 0.321, p = 0.309) significantly correlated with Year 3 language outcome. However, three other Year 1 MABC-2 scores were positively correlated with Year 3 language scores: the Total (r = 0.638, p = 0.025), Balance (r = 0.614, p = 0.034), and Manual Dexterity (r = 0.670, p = 0.017) scores.

Among the TD group, Year 1 SPELT-3, FVMC, and NRT scores were positively correlated with Year 3 language outcome (SPELT-3 r = 0.873, p <0.0001; FVMC r = 0.577, p = 0.015; NRT r = 0.564, p = 0.036), but none of the other predictor variables were correlated with language outcome in this group. Table S2 of the Supplemental Materials details the correlations between the predictor variables and Year 3 language outcome, as well as the correlations between the predictor variables at Year 1, for the DLD group and the TD group.

To assess the relative contribution of early language, speech-sound, and motor measures in predicting language outcome, a hierarchical regression was performed for both groups (DLD n = 12; TD n = 14). Neither multicollinearity nor heteroskedasticity were concerns for the DLD or TD groups; all predictors were included in each model. For both groups, the overall model accounted for a significant proportion of variance in language outcome, DLD: R² = 0.877, F(5,6) = 8.551, MSE = 34.089, p = 0.011; TD: R² = 0.860, F(5,8) = 9.815, MSE = 35.362, p = 0.003.

For the DLD group, the hierarchical regression analysis revealed an increase in the proportion of variance explained in language outcome by including MABC-2 Total in addition to the language and speech-sound measures (change in R² = 0.403, F(1,6) = 19.633, p = 0.004). That is, Year 1 fine/gross motor scores uniquely explained about 40% of the variation in Year 3 language scores over and above the Year 1 language and speech-sound scores.

Conversely, for the TD group, the hierarchical regression analysis revealed no increase in the proportion of variance explained in language outcome by including additional predictor variables beyond language. In other words, the language measures alone (SPELT-3 and FVMC) uniquely explained about 76% of the variation in Year 3 language scores, R² = 0.762, F(2,11) = 17.653, p <0.001 and the inclusion of subsequent speech-sound and fine/gross motor measures did not significantly increase the proportion of variance explained. The R², change in R², and the relevant corresponding statistics from Blocks 1 through 3 are presented in Table II for the DLD group and the TD group.

Table II. Model summary for hierarchical multiple regression of language, speech-sound, and motor predictors for each group (DLD, TD).

Model	Predictors	R^2	Std. error	Change in R^2	F change	df1	df2
DLD group
1	Year 1 FVMC	0.143	12.577	0.143	0.754	2	9
	Year 1 SPELT-P2
2	Year NRT Total	0.474	11.173	0.331	2.202	2	7
	Year 1 BBTOP
3	Year 1 MABC total	0.877	5.839	0.403	19.633**	1	6
TD group
1	Year 1 FVMC	0.762	6.602	0.762	17.653**	2	11
	Year 1 SPELT-3
2	Year NRT total	0.828	6.217	0.065	1.703	2	9
	Year 1 BBTOP
3	Year 1 MABC Total	0.860	5.947	0.032	1.836	1	8

Notes. Each block subsumes the previous block’s predictors, as well.

Dependent variable: language outcome—Year 3 CELF-4 CLS SS.

*p<0.05; **p<0.01.

The unstandardised regression coefficients and the corresponding statistics for each of the blocks are presented in Table S3 of the Supplemental Materials. For the children with DLD, none of the relationships in the first two blocks were significant; in the final block, there was a statistically significant relationship between language outcome and Year 1 BBTOP (p <0.05), NRT (p <0.05), and MABC-2 Total (p <0.01). Unexpectedly, the relationship between Year 1 BBTOP and Year 3 language was negative. For the children with TD, in each block, the only significant relationship was between Year 1 and Year 3 language (p <0.01).

Individual developmental trajectories

In the preceding section, we demonstrated that certain early fine/gross motor measures in combination with early language measures improved the prediction of language outcome in children with DLD as a group. Given our small sample size, we could more closely consider individual developmental profiles to gauge the clinical relevance of early fine/gross motor skills in predicting language outcome in individual children identified as having DLD at Year 1. Table III presents the language and fine/gross motor scores for each of these children. Early fine/gross motor abilities as measured by the MABC-2 Total were significantly weaker in children whose language impairment persisted into Year 3 than in children who scored within the typical range at Year 3 (t(12) = −2.237, p = 0.0450). Notably, Year 1 language scores in children whose language impairment persisted into Year 3 were not significantly lower than in children who ultimately scored within the typical range at Year 3 (t(12) = 0.114, p = 0.911), highlighting the important contribution of early motor deficits in predicting language outcome.

Table III. Individual profiles of motor and language scores in children whose language impairments did (participants 1–6) and did not (participants 7–14) persist to Year 3.

Participant	MABC-2 total	SPELT-P2	CELF-4 CLS	Participant	MABC-2 total	SPELT-P2	CELF-4 CLS
	Timepoint 1^a	Year 1	Year 3 outcome		Timepoint 1^a	Year 1	Year 3 outcome
DLD1	8	86	67	DLD7	4^a	64	87
DLD2	4^a	86	73	DLD8	7^a	65	90
DLD3	6^a,b	59	75	DLD9	9	67	93
DLD4	9	81	84	DLD10	10	87	94
DLD5	5^a	70	85	DLD11	17	68	94
DLD6	7^a	65	85	DLD12	12	79	102
				DLD13	14	76	114
				DLD14	11^b	85	118
Average	6.5 (1.87)	74.5 (11.47)	78.17 (7.60)	Average	10.5 (4.04)	73.88 (9.14)	99 (11.38)

Notes. Unhighlighted cells (white) indicate scores within the typical range; cells highlighted in grey reflect scores below the typical range.

^aAll six participants with impaired motor (MABC-2 Total) at Timepoint 1 demonstrated impairments in language, speech, or both language and speech at Year 3 (Year 3 BBTOP scores for these six participants: DLD2 = 84; DLD3 = no speech outcome data available; DLD5 = 78; DLD6 = 80; DLD7 = 77; DLD8 = 81).

^bIn these two instances (DLD 3 and DLD 14), data for Motor Timepoint 1 were obtained from Year 2 of the study; DLD 3 and DLD 14 did not complete the MABC-2 at Year 1, and therefore, these scores reflect the earliest timepoint at which they completed the task. These Year 2 data points were used only in the visualisation of individual trajectories, not in any of the other analyses presented. That is, all correlation and regression analyses presented throughout the manuscript included data from Year 1 and Year 3 timepoints only.

Articulatory variability as a predictor of language outcome

An independent samples T-test showed no differences in STI between children with DLD (M = 25.25, SD = 6.50) and TD (M = 22.16, SD = 6.12), t(23) = 1.209, p = 0.239. Correlation (DLD: r = 0.360; TD: r = −0.043) and regression analyses (DLD: R² = 0.129; SE = 13.673; TD: R² = 0.002; SE = 13.122) revealed no relationship between Year 1 STI and Year 3 language outcome for children in either group, shown in Tables S4 and S5 of the Supplemental Materials.

Discussion

In this small longitudinal study, among preschoolers who met the standard diagnostic criteria for DLD (and SLI), fine/gross motor scores—not language or speech scores—predicted language outcome two years later. That is, preschool language ability alone was not a significant predictor of school-age language outcome. Moreover, though early morphosyntactic and speech-sound deficits were often co-morbid, speech-sound deficits did not predict poor language outcome in our sample. Rather, among this sample of preschoolers diagnosed with DLD, deficits in certain fine/gross motor measures predicted persistent language impairment.

By contrast, for children with TD, only the language scores at Year 1 predicted language outcome scores. Speech-sound, speech-motor, and fine/gross motor scores did not. Though children with TD by definition scored within the typical range on language measures at Year 1, their CELF-4 CLS scores at Year 3 did have variability (M = 108.79, SD = 12.46, Range: 90–126). This study suggests that there may be a unique relationship between fine/gross motor and language skills in children with DLD that does not extend to typically developing children.

As is consistent with the diagnostic criteria for DLD/SLI, speech-sound and fine/gross motor skill were free to vary, though DLD was the primary diagnosis of all individuals in this group and they did not have known diagnoses of DCD or motor speech impairments (e.g. childhood apraxia of speech). Though the participants met the rigorous diagnostic criteria for SLI and we ruled out overt structural deficits through an OME, we do not yet fully understand how speech and motor domains interact with DLD—the rationale for this work. For example, it remains controversial whether an OME can rule out oral motor deficits (e.g. Strand, McCauley, Weigand, Stoeckel, & Baas, 2013). Therefore, it is possible that some individuals also had clinical or subclinical speech motor and/or fine/gross motor deficits (or neurodevelopmental immaturity) of which we were unaware.

It is also possible that response to outside intervention accounts for some portion of the improvement in language scores longitudinally. Our limited information regarding outside intervention precludes us from ruling out this possibility. However, severity (which likely relates to intervention frequency/duration) did not predict outcome, and all children received at least some intervention; therefore, outside intervention likely does not explain the results. Instead, we explore how the results may align within a framework of complex sequential learning processes.

DLD has classically been viewed as a deficit specific to linguistic processes, but alternative theoretical explanations of language impairment implicate specific extralinguistic deficits in the profile. In the current longitudinal study, we found that considering measures beyond language alone enhanced the prediction of language outcome in preschoolers with DLD. This aligns with accounts that implicate particular extralinguistic domains in DLD and, though limited by the small sample size, provides evidence that some co-morbidities (e.g. motor deficits) may in fact be mechanistically tied to the disorder. Though we do not argue that impaired sequential learning fully accounts for DLD, we do propose that it appears to be one important mechanism underlying certain language and motor skills.

In our study, early standardised language scores did not correlate significantly with language outcome among preschoolers with DLD at Year 1, although poor language skills at this age were indeed a risk factor (evidenced by the fact that no children with TD at Year 1 had poor language outcomes at Year 3). Though early speech-sound and nonword repetition deficits often co-occurred with early language impairment at age 4–5, this combination of deficits alone did not correlate with language outcome at age 6–7. This null finding might reflect our small sample size; however, it may also serve as preliminary evidence that articulation deficits (i.e. one or few sound omissions and/or substitutions) assessed by simple single word tasks, though often co-morbid with language impairments in preschoolers with DLD, may not be mechanistically related. By contrast, more complex tasks that obligate the organisation of phonological sequences, as well as more complex motor speech deficits (e.g. childhood apraxia of speech), may be more directly tied to language, as opposed to speech-sound disorder. Articulatory variability in productions of agent + action phrases differentiated children with DLD from those with TD and from those with speech sound disorder (SSD) without language impairment (Vuolo & Goffman, 2018), and children with DLD had more difficulty organising syllable sequences than producing individual sounds in a novel word form learning task (Benham et al., 2018). Single word articulation tests, such as the BBTOP, that afford just one snapshot of a production and are not designed to capture variability in productions, may be less predictive of language outcomes.

Findings that suggest articulatory variability as an important deficit in DLD also help interpret the initially unexpected result that, in the final block of the hierarchical regression for children with DLD, Year 1 BBTOP scores were negatively related to Year 3 language outcome. Again, this may be due to our relatively small sample size; however, it may also reflect these limitations in the BBTOP as a predictor, or the challenges of differential diagnosis at Year 1. Grammatical omissions may be driven by a weak phonetic inventory and possible SSD, and therefore, in some cases the Year 1 speech score may reflect deficits aligned with SSD, perhaps explaining the negative correlation with language outcome. More evidence is needed to decide whether, and which of, these potential explanations accounts for the negative relationship.

Importantly, language outcomes were poorest in children whose early language deficits co-occurred with fine/gross motor deficits. The strong correlation between MABC-2 scores and language outcome might be surprising given that sequence learning is not necessarily implicated in general motor tasks. However, it is notable that only the Balance and Manual Dexterity tasks, not the Aiming and Catching tasks, were significantly correlated with language outcome. This breakdown is consistent with the profile of motor weaknesses reported in children with DLD (e.g. Hill, 2001; Sanjeevan & Mainela-Arnold, 2019) and aligns with our theoretical account of DLD in which cognition underlies both language and motor deficits.

For example, while the Manual Dexterity tasks are global measures not designed to specifically evaluate sequential patterning, tasks in this subtest, such as threading beads and posting coins, do obligate manually sequencing elements. Furthermore, though balance does not necessarily implicate sequential processing, it does rely on higher-order cognitive processes and is related to cognitive load (e.g. Champion, Rose, Payne, Burns, & North, 2014; Haddad, Ryu, Seaman, & Ponto, 2010). While the Aiming and Catching subtest does involve sequential movements, this subtest may be more influenced by experiential factors (e.g. practice and experience playing catch outside the laboratory setting) than the Manual Dexterity and Balance subtests, perhaps leading to more rote movement patterns than novel sequential movements. Additionally, the protocol for the Aiming and Catching task is fairly unconstrained: When throwing, there are no restrictions on the manner in which the beanbag is tossed (e.g. underhand, overhand, dominant hand, non-dominant hand), and any part of the beanbag may land on any part of the target mat. When catching, children may trap the beanbag against their body. As such, less emphasis is placed on the coordinated sequential movements used to throw and catch. By contrast, for example, in the trail drawing task on the Manual Dexterity subtest, performance is measured down to the millimetre, requiring more finely coordinated sequential movements for successful completion of the task. It is clear, however, that a more nuanced approach to studying complex rule-based sequencing is critical for better understanding its role in DLD (Gerken, Plante, & Goffman, 2021).

To more directly measure sequential patterning, we included an experimental sentence repetition task. We found no group differences on this task in the STI, a measure of articulatory variability, at Year 1, nor was increased articulatory variability predictive of poor language outcome at Year 3. These null findings are likely due to the simplicity of the repetition task. Performance differences between children with TD and those with DLD on procedural learning tasks are most often observed on tasks that have high sequential complexity (e.g. Hill, 2001; Saletta et al., 2018; Sanjeevan & Mainela-Arnold, 2019; Vuolo et al., 2017). The rote nature of this sentence repetition task may have enabled children to rely on the declarative system rather than sequential learning mechanisms, resulting in similar performance across groups. Although our original experimental task included a primed sentence generation condition in which children heard a sentence and generated a syntactically similar one with new content words (Saletta et al., 2018), the majority of children with DLD could not complete this task at Year 1 due to its difficulty level. This precluded an analysis of STIs in the higher complexity condition. At later timepoints, the DLD group was significantly more variable on this sentence generation task than the TD group (Saletta et al., 2018). We hypothesise that a group difference in articulatory variability would also be observed in the preschool years, and may predict language outcome, in a more complex task. This is an important future research direction to understand whether certain motor deficits are tied to DLD through shared impairments in sequential learning mechanisms.

Clinical implications

This longitudinal study elucidates the complex profile of DLD, highlighting the interactivity among the language, speech, and motor domains and suggesting that some or all of these areas may be core deficits of DLD. This complexity may help explain the difficulty in accurately identifying children with DLD at a young age based on language-specific measures only (e.g. Rudolph et al., 2019). This study reveals that preschool-aged children with DLD who have weaknesses in both language and fine/gross motor domains are more likely to have language deficits that persist into the early school years than preschoolers who have language deficits only. We suggest that clinicians be cognisant of the role that early motor deficits may play in the prediction of persistent language impairment and consider fine/gross motor abilities when evaluating for language impairment. Deficits in fine/gross motor skills can be recognised before overt morphosyntactic deficits become apparent; awareness of the interconnectivity between the motor and language domains may facilitate earlier intervention services for young children at risk for or demonstrating language impairment. Additionally, continued research investigating the role of variability in DLD is important, and we urge researchers and clinicians to attend to variability in both real-word (e.g. via the Diagnostic Evaluation of Articulation and Phonology, Dodd et al., 2006) and nonword production tasks. The precise nature of the motor deficits implicated in DLD remains unclear; future research should determine the specific motor tasks that may be most predictive of language outcome, particularly those with complex sequencing demands.

Acknowledgements

We thank the study participants and their families, and Dr. Pat Deevy for her assistance with recruitment and assessment.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed at http://10.1080/17549507.2021.1998629.

Additional information

Funding

This research was supported by the National Institute on Deafness and Other Communication Disorders Grants R01 DC04826 and DC016813 awarded to Lisa Goffman.

Notes

1 Though the category of DLD may be broader than that of SLI, we refer to each fairly interchangeably throughout this work. We use the term SLI when that reflects the perspective of the researcher to whom we are referring, and when it more aligns with a particular point we are discussing. Because we investigate domains beyond language, we conservatively approach the standard inclusion criteria for SLI (Leonard, 2014) to ensure alignment with classic studies.