Speech motor development is sequential; articulators mature at different rates. In early infancy, jaw movements are rudimentary and mostly limited to simple opening and closing movements reflecting an undeveloped ability for precise force control (e.g., Davis and MacNeilage, 1995; Green et al., 2000, 2002; Kent, 1992; Locke, 1983; Namasivayam et al., 2020; Nip et al., 2009; but see Diepstra et al., 2017; Giulivi et al., 2011). During the first year, there is minimal interaction between the lips and the jaw, and the tongue shows little capacity for elevation away from the mandible (Buhr, 1980; Kent, 1992; Otomo and Stoel-Gammon, 1992). By around age two, as coordination between the laryngeal system and oral structures (including the jaw) improves, children begin to exhibit voicing contrasts (Green et al., 2002; Grigos et al., 2005; Yu et al., 2014). At this stage, strong interlip coupling is evident; between ages two and three, this coupling gradually differentiates, allowing the upper and lower lips to move independently, which is critical for the development of labiodental fricatives like /f/ and /v/ (Green et al., 2000, 2002; Green and Nip, 2010; Nip et al., 2009; Stoel-Gammon, 1985).

From ages two to six, lip movements become more finely tuned, and by age three, the tongue starts to operate with increasing independence from the jaw, enabling more accurate anterior–posterior movements (Donegan, 2013; Kent, 1992; Otomo and Stoel-Gammon, 1992; Smit et al., 1990; Wellman et al., 1931). Between three and five years, enhanced coordination between the tongue and jaw emerges, which is essential for articulating more complex sounds (Kent, 1992; McLeod and Crowe, 2018). Because the tongue functions as a hydrostatic organ, achieving fine articulatory coordination depends on both its maturation and exposure to language specific gestural contrasts (Green and Wang, 2003; Kent, 1992; Nittrouer, 1993; Noiray et al., 2013). As the tongue’s various components mature, more intricate sound such as rhotacized vowels and complex fricatives begin to emerge. Ultimately, speech motor variability decreases and stabilizes between the ages of seven and twelve, as coordination among the lips, jaw, and tongue becomes increasingly consistent and efficient (Cheng et al., 2007; Nittrouer, 1993; Nittrouer et al., 1996, 2005; Smith and Zelaznik, 2004; Zharkova et al., 2011; see Figure 1 in Namasivayam et al., 2020).

To summarize, the progression of oral articulatory skills occurs along a variable timeline. The coordination between the lips and the jaw reaches maturity before that of the tongue with the jaw or the independent movement of various tongue segments (Cheng et al., 2007; Terband et al., 2009). Furthermore, speech motor control emerges in a hierarchical, step-by-step, and non-uniform manner, evolving over a prolonged period, which in turn plays a crucial role in acquisition and accurate production of speech sounds (Smith and Zelaznik, 2004; Whiteside et al., 2003). For a comprehensive look at speech motor development see Namasivayam et al. (2020).

Infants across languages initially produce a limited range of speech sounds and follow a similar order in mastering them, often employing analogous simplification strategies (Green et al., 2002; Locke, 1983; Preisser et al., 1988; Sander, 1972; Stoel-Gammon, 1985). This developmental pattern indicates that emerging speech motor skills most likely influence speech sound acquisition and production (Green et al., 2002). Young children (less than 2 years of age) successfully produce speech sounds that can be effectively formed using the mandible as the primary mover (e.g., /b/), and are less able to produce those that tend to be associated with graded lip control (e.g., /f/).

Further, bilabial stops (/p/, /b/) are highly represented in early phonemic inventories as they can be produced using relatively ballistic jaw movement without active contribution from the lips or tongue (Kent, 1992; MacNeilage and Davis, 1990). On the other hand, sounds like the labiodental fricative /f/, which require fine, independent control of the lower lip and jaw, typically emerge later around 2.5 years of age and are mastered around age 4 (Sander, 1972; Stoel-Gammon, 1985). The preference for jaw-supported sounds in early speech may be attributed to the biomechanical stability of the mandible, a single bone that articulates bilaterally with the temporal bones and is supported by a symmetrical network of muscles (i.e., jaw depressors and elevators), thereby limiting extraneous movements (i.e., limited degrees of freedom in movement directions not related to speech) and providing a stable base for speech (Green et al., 2002; Shiller et al., 2002; Van Lieshout, 2015).

Speech sounds that require even greater finer articulatory control, such as fricatives (/s/), affricates (/t∫/, /dʒ/) laterals and rhotics (/l/, /ɹ/), are mastered later (4+ years) due to the increased demand for independent lip and tongue movements (Kabakoff et al., 2021, 2023; Lin and Demuth, 2015; MacNeilage and Davis, 1990). The control over the interdigitated muscular layers that compose the highly deformable hydrostatic organ, like the tongue, requires relatively greater fine-force regulation and sustained effort over time in comparison with those produced with a more ballistic closure (Abbs et al., 1984; Kabakoff et al., 2021, 2023; Smith and Kier, 1989). These findings indicate that early limitations in speech motor coordination can shape the order in which phonemes are acquired. Green et al. (2000, 2002) highlight that young children’s sound production is restricted by factors such as a reliance on jaw movement, limited coordination between the lips and jaw, limited lip control, and insufficient independent movement of the upper and lower lips, and various functional components of the tongue operating with increasing independence from the jaw (Namasivayam et al., 2020).

Contrary to the above view implicating a stronger role of speech motor development in speech sound acquisition, some researchers have argued that cross-linguistic differences in phoneme error patterns are best explained by factors such as functional load and phonological saliency (Dodd, 2014; Hua and Dodd, 2000). Hua and Dodd (2000) argued that the biological and articulatory constraints observed in young children cannot fully account for these cross-linguistic variations. For instance, they note that the infrequent occurrence of alveolo-palatal affricates in Putonghua (Modern Standard Chinese) and the earlier emergence of affricates in that language compared to English cannot be solely attributed to the frequency of these phonemes in the language or to inherent articulatory constraints (Dodd, 2014). Although the prominence of a phoneme in a child’s native language is important, recent acoustic and transcription findings by Ma et al. (2022) reveal that even within the realm of affricate acquisition in Putonghua language, those that involve the tongue body tend to be mastered earlier than the more complex alveolar and retroflex affricates. This evidence supports the oromotor maturation hypothesis, suggesting that children more readily control the muscles required for elevating the tongue body owing to its earlier development than those needed for raising the tongue tip (Kent, 2021; Li and Munson, 2016; Ma et al., 2022).

Collectively, these findings indicate that early phonological and speech sound development is shaped by several factors, including inherent neuromuscular organization, the spatial and temporal demands of articulating specific phonemes, and influences from the ambient language (Dodd, 2014; Hua and Dodd, 2000; Ma et al., 2022; Namasivayam et al., 2020). Additionally, studies that rely exclusively on auditory-perceptual transcription without instrumental data (e.g., Hua and Dodd, 2000) may have overlooked some of these subtle contrasts due to adults’ categorical perception biases (Kent, 1996; Ma et al., 2022; Meyer and Munson, 2021; Mowrey and MacKay, 1990; Namasivayam et al., 2020). These insights reveal that the process of acquiring speech sounds is more complex than once thought, reflecting a dynamic interaction among cognitive development, ambient language exposure, and oromotor maturation (Green and Nip, 2010; Kent, 2021; Ma et al., 2022; Namasivayam et al., 2020; Nip et al., 2011).

The connection between speech motor development and speech sound errors in children has received limited attention, likely due to the dominance of psycholinguistic models like Dodd’s Model of Differential Diagnosis (MDD; Bradford and Dodd, 1994; Broomfield and Dodd, 2004; Dodd, 2014), which traditionally attribute these errors to phonological factors. Dodd and colleagues integrated the psycholinguistic approach which focuses on input/output processing and internal representations with a clinical descriptive approach based on observed speech errors to develop the foundation for the MDD (Broomfield and Dodd, 2004; Dodd, 2014). Unlike purely developmental models such as Natural Phonology (Stampe, 1969), Dodd’s MDD emphasizes classifying SSDs based on the nature and consistency of error patterns. Within this framework, subgroups are identified according to surface (observed) error patterns, which are believed to reflect distinct underlying deficits in the speech processing chain, including perceptual, cognitive-linguistic, and articulation skills. According to the MDD framework, if a child exhibits phonological error patterns common among typically developing children, albeit with a slight delay, these errors are viewed as part of normal development, and the child is classified as having a “phonological delay.” In contrast, if the phonological error patterns are uncommon, i.e., occur <10% of the time in typically developing children, they are considered atypical, and the child is identified as phonologically disordered (Dodd, 2014).

The MDD model for classifying SSD subtypes is implemented via the Diagnostic Evaluation of Articulation and Phonology test (DEAP; Dodd et al., 2002), which identifies and evaluates ten typical and four atypical phonological error patterns in children. The typical phonological patterns include gliding (e.g., “rabbit” → “wabbit”), vocalization of liquids (e.g., “table” → “tabo”), deaffrication (e.g., “chair” → “sair”), cluster reduction (e.g., “spoon” → “poon”), fronting (e.g., “car” → “tar”), weak syllable deletion (e.g., “banana” → “nana”), stopping (e.g., “fish” → “pish”), prevocalic voicing (e.g., “pig” → “big”), postvocalic devoicing (e.g., “dog” → “dok”), and final consonant deletion (e.g., “cat” → “ca”). In contrast, atypical phonological patterns include backing (e.g., “tap” → “cap”), consonant harmony (e.g., “dog” → “gog”), medial consonant deletion (e.g., “ladder” → “la-er”), and palatalization (e.g., “sip” → “ship”). Within the MDD, Dodd (2014) proposes that phonological error patterns typically stem from several sources. These include cognitive-linguistic factors, such as difficulties in learning the phonological rules of a language, or an unstable phonological system. Errors may also arise from anatomical issues (e.g., cleft lip/palate) or muscle function impairments (e.g., Childhood Dysarthria). Additionally, some articulation difficulties, like lisps, occur without an identifiable cause and are believed to result from mislearning the mapping between perceptual and articulatory systems.

In contrast to Dodd’s MDD, linguistic theories such as Natural Phonology (Stampe, 1969), Grounded Phonology (Archangeli and Pulleyblank, 1994), and Articulatory Phonology (Browman and Goldstein, 1992; Namasivayam et al., 2020; Van Lieshout and Goldstein, 2008) all explicitly acknowledge the influence of phonetic and articulatory factors on phonological processes, a perspective not emphasized in Dodd’s MDD. These theories recognize that ease of articulation and perceptual clarity play significant roles in shaping sound patterns across languages. While Natural Phonology focuses on developmental processes and innate simplification strategies in speech, Grounded Phonology adopts a more formalist, system-wide approach, integrating phonetics and phonology within constraint-based frameworks, often aligned with Optimality Theory. It explains phonological markedness through universal phonetic motivations, suggesting that certain sounds are avoided across languages due to articulatory complexity or perceptual ambiguity.

Articulatory Phonology (AP), as outlined by Namasivayam et al. (2020), provides a dynamic approach to understanding speech sound disorders in children, focusing on the role of articulatory gestures which control coordinated movements of the speech organs in both typical and disordered speech production. This framework suggests that SSDs often stem from disruptions in the planning and execution of these gestures, leading to observable speech errors. Importantly, Namasivayam et al. (2020) posits that many of these errors may not solely represent deficits but also serve as compensatory strategies aimed at enhancing speech motor stability. Such adaptations include increased movement amplitude, slower speech rates, tongue bracing, intrusion gestures, cluster reduction, and deletions at the segmental, gestural, or syllabic levels. Additionally, increased phase lag between articulatory movements has been observed as a means of improving motor stability and intelligibility, particularly in speakers with less developed speech motor skills (Fletcher, 1992; Namasivayam et al., 2020; Van Lieshout et al., 2004). This integrative framework combines elements of speech perception, motor execution, and neural control, offering a comprehensive perspective for understanding and addressing SSDs. While Grounded Phonology focuses on explaining phonological patterns through universal phonetic constraints and articulatory ease, Articulatory Phonology emphasizes the real-time coordination and motor processes involved in speech production. This allows for a more detailed exploration of how breakdowns in gestural coordination contribute to speech errors, highlighting the complex interplay between motor control and phonological output in both typical and disordered speech (Namasivayam et al., 2020; Van Lieshout and Goldstein, 2008).

Notably, within the AP model articulatory gestures are defined at an abstract cognitive level by parameters related to constriction location and degree and by their temporal organization. They function as the speech system’s basic phonological primitives. Gestures combine, overlap, or are withheld to create contrast at the level of segments, words, and larger utterances. Linguistic phonological contrast can arise from the presence/absence of a gesture or from parametric differences within the same gesture. For example, “ban” differs from “bad” by the addition of a velic-lowering (nasal) gesture; “bad” versus “pad” is distinguished by a laryngeal gesture supporting voicing in the former; and “bus” versus “but” reflects a difference in tongue-tip constriction degree (narrow, frication-supporting aperture for /s/ vs. complete closure for /t/). Each gesture has an internal temporal profile with landmarks (onset, target, release), and the alignment of these landmarks across gestures yields the observable segmental structure and higher-level units. These patterns of gestures and their timing relations are typically schematized as a gestural score (Browman and Goldstein, 1990, 1992; Van Lieshout et al., 2008). Thus, within the AP framework, gestures are not defined at the articulatory level (level of execution) but rather at the cognitive stages of speech production (e.g., Goldstein et al., 2006, 2007). In this account, those gestures and subsequent stages in the model substitute for theoretical concepts of phonemes and features as used in more traditional psycholinguistic theories.

Let’s briefly look at potential mechanisms for a speech motor basis of so-called “phonological” process errors.

Given prior evidence on speech motor development, young children’s motor limitations are likely to contribute to systematic speech errors. In early speech development the reliance on ballistic jaw movements may mean that certain ‘phonological’ error patterns, such as substitutions and omissions, are more likely to occur in sounds requiring graded control (Kent, 1992; Tobin, 1997). In two-year-olds, the upper and lower lips typically move in unison with the jaw rather than independently. At this stage, a child might correctly produce /f/ only in jaw-supported contexts such as in transitions from an open-to-closed position (e.g., “off”) or closed-to-open (e.g., “fan”) but struggle with words like “fit” or “fish”, where the jaw remains in a stable, high-vowel position. This limitation can result in substitution errors, such as producing “pish” for “fish”. Difficulties with labiodental fricatives in early speech may also arise from the child’s inability to sustain airflow and control lip movements with the precision required for accurate production (Namasivayam et al., 2020; Tobin, 1997).

Similarly, children under the age of three, especially those with SSDs, often exhibit undifferentiated tongue gestures, i.e., limited independent tongue-tip elevation (Gibbon, 1999). As a result, alveolar consonants are frequently produced using jaw-supported tongue elevation (often observed as the jaw and tongue moving in the same direction, i.e., in-phase movements), creating what is known as jaw-compensated speech. In this synergy between the jaw and tongue tip, the jaw aids in achieving the necessary tongue-tip elevation for alveolar sounds. Thus, it is not surprising that the SSD population also has a high prevalence of (anterior or lateral) jaw sliding, likely representing an adaptive jaw response to reduced tongue control (Mogren et al., 2020, 2022; Namasivayam et al., 2013, 2020; Terband et al., 2013).

Such an adaptive jaw response to tongue movement limitations has also been reported in the adult neurodegenerative literature (Rong and Green, 2019). In the early stages of bulbar Amyotrophic Lateral Sclerosis (ALS), as tongue control deteriorates and speech intelligibility decreases, jaw-supported tongue movements have been observed as a compensatory strategy to counteract this decline in intelligibility. However, as bulbar ALS progresses, this adaptive jaw strategy diminishes, leading to a further reduction in speech intelligibility (Mefferd and Dietrich, 2020; Rong and Green, 2019). Furthermore, while the jaw-compensated speech strategy may help maintain intelligibility at the word level, it can reduce clarity in connected speech due to slower movements of the jaw (Green et al., 2000, 2002).

Thus, common phonological error patterns, such as cluster reduction, fronting, and stopping, can be attributed to issues with the ongoing refinement of speech motor control and the development of articulatory coordination. Additionally, errors that affect syllable and word shapes factors known to significantly impact speech intelligibility (Hodge and Gotzke, 2011; Osberger and McGarr, 1982) are likely rooted in the constraints of an immature oromotor system, as well. Next, we show how motor constraints may underlie DEAP “phonological” processes (Dodd et al., 2002), elaborated in Namasivayam et al. (2020).

To what extent are specific speech motor control limitations (e.g., lateral jaw sliding, limited tongue tip elevation from the jaw, inadequate jaw–lip integration) associated with, and predictive of, the occurrence and type of phonological error patterns (as identified by the DEAP) in preschool-aged children with moderate-to-severe SSDs?

This study analyzed pre-treatment data from 48 children (mean age: 48 months, SD = 11) drawn from a larger randomized controlled trial on speech motor intervention (Namasivayam et al., 2021a; Namasivayam et al., 2021b). Participants were recruited from community-based healthcare centers in Mississauga, Toronto, and Windsor, Ontario, Canada. Demographic details are presented in Table 1. Children were eligible if they met the following criteria: (1) age between 3 and 10 years, (2) English as the primary language spoken at home (language background data collected did not extend beyond the English-speaking requirement; we did not gather detailed information on additional languages the child may have been exposed to or using), (3) diagnosis of moderate-to-severe SSD, specifically an SMD subtype, based on features reported in the precision stability index (Shriberg et al., 2010; Shriberg and Wren, 2019), (4) normal hearing and vision (as confirmed by parent reports and school records), (5) Primary Test of Nonverbal Intelligence score at or above the 25th percentile with a standard score of ≥90 (Ehrler and McGhee, 2008), (6) receptive language standard score of ≥78 on the Clinical Evaluation of Language Fundamentals (Semel et al., 2003, 2004), (7) no restrictions on expressive language scores, (8) age-appropriate social skills, (9) age-appropriate play skills, (10) readiness for therapy: presence of intentional communication, (11) readiness for therapy: ability to imitate, (12) behaviourally ready for therapy and (13) presence of at least four motor speech limitation indicators as identified in the motor speech checklist (item 13–21; See Table 2 below; Namasivayam et al., 2013; Namasivayam et al., 2021a; Namasivayam et al., 2021b).

Children were excluded if they exhibited any of the following: (1) signs of global motor involvement (e.g., cerebral palsy), (2) more than seven out of 12 indicators on a CAS checklist (American Speech-Language-Hearing Association, 2007; cutoff from Namasivayam et al., 2015), (3) feeding difficulties, drooling, or oral structural/resonance issues, and (4) a diagnosis of autism spectrum disorder. A licensed SLP conducted all assessments for inclusion and exclusion criteria. The study received approval from the University of Toronto’s research ethics board (Protocol 29142), with additional approvals from participating clinical sites.

All assessment and intervention sessions were recorded in both video (JVC Everio GZ-E220 HD, 1,920 × 1,080 resolution) and audio (Zoom H1 Ver 2.0, 16-bit at 44.1 kHz) formats for inter-rater reliability analysis. κ-Statistics was used to measure agreement, with values categorized as poor (<0), slight (0.2), fair (0.4), moderate (0.6), substantial (0.8), and almost perfect (1) (Viera and Garrett, 2005). The κ coefficient, calculated from 20% of the data by blinded SLPs, averaged 0.73 (substantial) for auditory-perceptual (International Phonetic Alphabet) transcriptions for the DEAP assessments (Dodd et al., 2002). The reliability of observing speech motor control deficits across different articulators using specialized probe words (Namasivayam et al., 2021a) ranged from fair to moderate: mandibular (0.52), labial–facial (0.57), lingual (0.63), and sequenced items (0.48). κ-statistics for intra- and inter-rater reliability (calculated on 20% of randomly sampled data) for the motor speech checklist was 0.65 and 0.54, respectively. Note: The SLP who completed the motor speech checklist at study inclusion used different assessment methods (spontaneous speech sample, articulation testing, and a live sample) than the SLP who later scored the checklist using recorded probe words. Consequently, inter-rater reliability was somewhat lower than intra-rater reliability, as expected.

All assessments were carried out by licensed speech-language pathologists in a quiet room, with age-appropriate decor and stimuli using standardized testing procedures reported in the literature (Namasivayam et al., 2013; Namasivayam et al., 2021a; Namasivayam et al., 2021b). Furthermore, samples were also excluded from the analysis if there were missing data in either phonological or speech motor control records. In total, data from 30 children were included in the analysis after exclusion.

We conducted Mutual Information Analysis, a Random Forest model, and additional correlational analyzes (Pearson’s r) to interpret the findings, along with one-way ANOVA and Kruskal-Wallis tests to assess statistical significance. Given the high linguistic diversity of the recruitment regions, it is likely that some children were bilingual or multilingual; however, subgroup analyzes examining the effects of linguistic background on motor control were not feasible due to the small sample size and the wide variety of potential language pairings.

Figure 1 illustrates the pair-wise mutual information heatmap graph between speech motor control limitation (x-axis) and phonological error pattern (y-axis), with warmer colors (red) indicating stronger associations and the number within each cell the mutual information (NMI) of the feature pair. Specifically, phonological errors were primarily associated with three types of speech motor control limitations: labial-facial control (items 14.1 and 14.2 in Table 2), lingual control (items 16.1 and 16.2), and, to a lesser extent, jaw control (item 13.3). These findings are generally consistent with those proposed by Green et al. (2000, 2002) who identified jaw control, integration of the jaw and lips, upper–lower lip movement independence, and tongue–jaw dissociation as critical for accurate speech sound production in young children. In contrast, other speech motor limitations, including Multi-plane Movements (items 17), General Speech Production Characteristics (items 18– 21), and Integration of Jaw and Lips (items 13), exhibited only limited association with phonological errors (NMI < 0.20).

When examining phonological error patterns, FCD (final consonant deletion), GLIDING, and STOPPING demonstrated fair to moderate associations with both Labial-Facial and Lingual control limitations (items 14.1, 14.2, 16.1, and 16.2; NMI > 0.30). Meanwhile, the CC REDUCTION (cluster reduction) error only showed stronger association with speech motor control item LIP FACE 14.1 (NMI = 0.40). Among all feature pairs, phonological error FCD and motor control limitation LIP FACE 14.1 exhibited the highest connection with an NMI value of 0.47 (moderate association). On the other hand, phonological errors such as DEAFF (deaffrication), PRE-VOCALIC VOICING, POST-VOCALIC DEVOICING, and VOCALIZATION OF LIQUIDS demonstrated weaker associations with motor control errors, with the NMI below 0.20 across most motor control categories. Additionally, atypical phonological errors also presented stronger associations with posterior movement-related lingual control errors (LINGUAL 16.2, NMI = 0.46). Lastly, Atypical Patterns, FCD, PRE-VOCALIC VOICING and STOPPING had fair associations with Jaw sliding (anterior/lateral slide; item Jaw 13.3) with NMI values around 0.25 and 0.26.

Figure 2 presents the mean absolute SHAP values for each phonological error feature, quantifying their overall importance in the model. Among the features, GLIDING demonstrated the highest importance, with a mean SHAP value of approximately 0.21. Other features, including Atypical Patterns Total, FCD, and CC reduction, also contributed although with lower SHAP values. Features such as PRE-VOCALIC VOICING, POST-VOCALIC DEVOICING, and FRONTING showed relatively less affect.

The SHAP swarm plot (Figure 3) provides further insight into the directional influence of these phonological errors on model predictions. In this plot, the x-axis represents the SHAP values, indicating the magnitude and direction of a feature’s impact on the model’s output. Positive SHAP values suggest that the feature increases the predicted score, while negative values indicate a decreasing effect. The y-axis lists the phonological error features ranked by importance. Each dot corresponds to a single data sample, showing the SHAP value for that feature in that particular instance. The color gradient represents the feature value, with red indicating higher values and blue representing lower values.

Interestingly, as shown in Figure 3, the GLIDING error exhibited a negative impact on the prediction for most instances, which distinguished it from the other phonological features. This suggests that, higher gliding occurrences are typically associated with lower speech motor control error scores. In contrast, most other features, such as FCD, CC REDUCTION, and atypical patterns, generally displayed a positive association with the model output, meaning that increased occurrences of these errors tend to elevate the predicted score.

Figure 4 presents two specific prediction examples to demonstrate this counter-directional relation between GLIDING and total speech motor error score. Here, SHAP waterfall plots were employed to detail how individual phonological features contribute to the predicted speech motor control error score. In each graph, the x-axis represents the model’s predicted motor score f (x), with E[f (x)] presenting the average prediction across all samples. The y-axis lists the phonological features with their corresponding counts. Each bar represents the SHAP value for a feature, indicating how much it increases or decreases the prediction. In Figure 4A, the participant had a higher GLIDING error count of 18 and a lower total speech motor error score of 12. Thus, GLIDING played a role in reducing the predicted speech motor error score with a negative SHAP (−0.23). Meanwhile, a lower GLIDING count of 1, as shown in Figure 4B, presented a positive influence on the prediction with a SHAP value of +0.23. On the other hand, other phonological errors in both examples, such as FCD, CC REDUCTION, and Atypical Patterns, exerted the same direction influence on the predicted motor score, i.e., higher/lower phonological error counts increasing/decreasing the prediction.

Since gliding affects the later-acquired, complex /r/ and /l/ sounds, we expected higher motor speech checklist scores (i.e. more speech motor errors) to correlate with more gliding. However, contrary to our predictions, greater gliding was observed with lower speech motor control error scores. To further investigate, we examined whether gliding errors were more prevalent in older children and those with fewer speech motor issues compared to younger children or those with more severe speech motor control deficits. To assess this, we analyzed data from the demographics Table 1 (Namasivayam et al., 2021a; Namasivayam et al., 2021b), focusing on Percent Consonants Correct (PCC) as a measure of speech severity (>85% = mild, 65%–85% = mild-moderate, 50%–64% = moderate-severe, <50% = severe; DEAP test, Dodd et al., 2002) and speech motor scores from the standardized Verbal Motor Production Assessment for Children (VMPAC, Hayden and Square, 1999). The VMPAC systematically evaluates the neuromotor integrity of the speech motor system in children with speech sound disorders. For this study, we used the oromotor control and sequencing subsections of the VMPAC. Raw scores from these sections were divided by their total possible scores and converted into percentage scores (ranging from 0 to 100), where lower scores indicate poorer speech motor control, and higher scores reflect better speech motor performance.

We computed Pearson correlation coefficients between GLIDING and both age in months and percent consonants correct (PCC), respectively, to assess whether gliding is more frequent in older or milder children. The results show a statistically significant positive relationship between Gliding vs. Age (r = 0.407, p = 0.023; Figure 5) and Gliding vs. speech severity (r = 0.480, r = 0.006; Figure 6). Also, the linear regression lines fitted with the data demonstrate significant positive slopes in both cases.

Next, we divided patients into groups based on PCC severity: severe (PCC < = 50), moderate-severe (50 < PCC < 65), and mild-moderate (PCC > = 65). As there was only one mild participant, the mild and mild-moderate groups were combined. A one-way ANOVA as well as a Kruskal-Wallis test was utilized to see if Gliding counts differ significantly across PCC severity groups. Note that Kruskal-Wallis test is a non-parametric version of the one-way ANOVA (no normality assumption). Figure 7 compares the distribution of gliding errors in these three PCC severity groups. There were significant differences in gliding errors of the three groups (ANOVA p = 0.023; Kruskal-Wallis p = 0.026).

We also compared the relative frequency of gliding errors to all phonological errors across different PCC severity groups. The goal was to determine whether the milder PCC group exhibited more gliding errors but fewer other phonological errors, while the severe PCC group had fewer gliding errors and more of other phonological error patterns. To assess this, we calculated the ratio of gliding errors to other phonological errors, referred to as the gliding ratio, across different PCC severity groups. Statistical significance was analyzed using one-way ANOVA and Kruskal-Wallis tests. As shown in Figure 8, the gliding ratio differed significantly between severity groups (ANOVA: p < 0.001; Kruskal-Wallis: p = 0.001).

Finally, we assessed the convergent validity of these findings by calculating Pearson correlation coefficients between gliding errors, total speech motor checklist scores, and the VMPAC standardized speech motor assessment (Hayden and Square, 1999). VMPAC scores were significantly negatively correlated with total speech motor checklist scores, r(30) = −0.45, p = 0.012, while they were significantly positively correlated with total gliding errors, r(30) = 0.59, p = 0.0006.