Early language skills and later internalizing and externalizing behaviours: associations through core and pragmatic language

Abstract

Background

Language skills have been associated with both internalizing (anxiety, depression, somatization) and externalizing (conduct problems, hyperactivity, aggression) behaviours. Studies investigating these links have primarily focused on core language skills, such as expressive and receptive language, with less attention to pragmatic language, the social use of language in everyday interactions. The present study examined associations between early expressive vocabulary and later internalizing and externalizing symptoms through both core and pragmatic language skills.

Methods

Using data from a large, prospective Canadian cohort study, structural equation models tested longitudinal associations between expressive vocabulary skills at age 2, core and pragmatic language at age 5, and internalizing and externalizing symptoms at age 8.

Results

Results indicated significant indirect associations between early expressive vocabulary and later internalizing and externalizing symptoms through both core and pragmatic language.

Conclusions

These findings highlight the importance of considering the role of both core and pragmatic language skills when examining associations with later behavioural symptoms.

Early childhood language development serves as a critical foundation for various developmental domains, including self-regulation and executive function [1–3], academic achievement [4–6], and social competence [7–9]. Language skills also have been linked to mental health, with meta-analyses indicating significant associations between language abilities and general psychological problems [10, 11] and between language skills and behavioural and emotional difficulties [12–14]. Despite this broad evidence base, results are not uniform across studies [15–17]. This variability raises the possibility that early language may not influence emotional and behavioural functioning directly, but rather through intermediary developmental processes. Furthermore, much of the research investigating the link between language skills and behavioural problems in children has focused on core language skills, including receptive and expressive language [18–22]. This leaves unanswered questions around the role that other aspects of language development, such as pragmatic language skills, may play in these associations. The current study addresses these gaps by investigating whether associations between early expressive vocabulary at age 2 and internalizing and externalizing symptoms at age 8 are mediated through children’s core or pragmatic language skills at age 5, drawing upon data from a large longitudinal prospective cohort study.

To begin, we first describe our constructs of interest. We use the term core language skills to encompass the basics of understanding and transmitting language, including semantics and syntax [23]. Pragmatic language skills, in turn, deal with the process of using language in a social context, comprised of skills such as turn-taking and staying on topic. Pragmatic language skills aid in social communication, friendship building, and understanding nuance and implied meanings [24]. We define externalizing symptoms as consisting of outwardly directed behaviours including hyperactivity, aggression, and conduct issues. Internalizing symptoms consist of inwardly directed difficulties including anxiety, depression, and somatization [25].

Core language

A large body of research has examined longitudinal associations between language skills and internalizing and externalizing problems in typically developing children, with studies yielding mixed findings [18, 19, 26–28, see 29, 30 for reviews]. For example, poorer receptive language skills during the preschool years predicted later internalizing and externalizing behaviours over a period of two years [28]. These associations also appear bidirectional, with behavioural difficulties in pre-kindergarten (e.g., externalizing, bullying, hyperactivity, internalizing) associated with poorer vocabulary outcomes by the end of the school year [19]. Similarly, some large-scale community-based cohort studies across diverse contexts have reported associations between early language and later internalizing or externalizing symptoms across childhood [15, 17, 20–22, 31–33]. For example, a longitudinal study of over 100,000 Norwegian children found bidirectional relations between language difficulties and internalizing problems from 18 months to 8 years [20]. Other cohort studies have reported similar associations between early receptive or expressive vocabulary and later emotional or behavioural functioning from toddlerhood through adolescence [22, 31–33], although these effects are often modest and inconsistent across language domains, developmental periods, and cohorts. For example, Thornton et al. [17] found that receptive vocabulary at age 5 predicted adolescent internalizing symptoms in only one of two cohorts sampled. Henrichs et al. [15] reported inconsistent associations across developmental periods, with expressive vocabulary delay associated only with internalizing problems at 18 months but predicting both internalizing and externalizing problems at 36 months of age. More recently, Ömercioğlu et al. [16] demonstrated that associations with externalizing problems varied by language domain, such that receptive language skills were associated with fewer conduct problems and lower levels of hyperactivity, whereas expressive language skills showed no significant relation with these behaviours.

To address the complex and sometimes inconsistent pattern of findings in the literature, Hentges et al. [13] conducted a meta-analysis examining associations between language skills and internalizing and externalizing problems in childhood, as well as factors that might moderate these relations. Drawing upon data from 139 studies, they found that lower language skills were associated with greater psychological difficulties, with similar small but significant effects observed for both internalizing (Hedges’ g = 0.23) and externalizing problems (Hedges’ g = 0.22). These associations were observed both in dimensional studies of community cohorts and in comparisons between language-impaired and typically developing children, suggesting that language difficulties may be a risk factor for emotional and behavioural difficulties across the population. This meta-analysis, however, examined language skills broadly and did not differentiate associations across specific language domains, nor did it consider pragmatic language abilities.

Pragmatic language

As pragmatic language encapsulates the social use of language in everyday interactions, it may provide a critical connection between early language abilities and later internalizing and externalizing problems. Pragmatic language skills refer to the use of language in social contexts, including turn-taking, topic maintenance, conversational repair, and interpretation of implied meaning [34, 35]. These skills directly support children’s participation in everyday social interactions, and therefore pragmatic language has been proposed as a unique predictor of socio-emotional and behavioural functioning [36]. Difficulties with pragmatic language may disrupt peer interactions, increase social conflict, and limit children’s ability to navigate social expectations, increasing risk for both internalizing and externalizing symptoms [37–39].

Research examining the relations between pragmatic language skills and internalizing or externalizing symptoms has largely focused on clinical populations, including children with autism spectrum disorder [ASD; 40], developmental language disorder [DLD; 41, 42], and attention-deficit hyperactivity disorder [ADHD; 43]. Across these groups, pragmatic language difficulties are well documented relative to typically developing children [42, 44–46], with lower pragmatic language skills in these groups consistently associated with greater internalizing and externalizing symptoms [44, 47]. Several clinical studies indicate that pragmatic language explains behavioural difficulties above and beyond core language skills, highlighting the unique contribution of pragmatic language difficulties to behavioural symptoms [44, 47, 48].

Research examining pragmatic language skills in relation to internalizing and externalizing behaviours in typically developing children remains limited. In a large population-based study of over 1,300 four-year old children, Ketelaars et al. [36] found that pragmatic language competence, assessed using a Dutch translation of the first edition of the Children’s Communication Checklist (CCC), showed strong concurrent associations with behaviour problems at 4 years of age. Pragmatic language difficulties were associated with externalizing behaviours, including hyperactivity and reduced prosocial behaviour, while associations with internalizing symptoms were present but more modest. Importantly, structural aspects of language (i.e., speech and syntax) did not account for unique variance in behavioural outcomes once pragmatic language abilities were considered. Relatedly, Camia et al. [49] found that school-aged children’s pragmatic difficulties—particularly stereotyped language and inappropriate context use—were associated with higher levels of both internalizing and externalizing symptoms. Finally, Ahmed et al. [50] reported that children who produced more coherent and relevant written language, which they viewed as marking stronger pragmatic language skills, exhibited fewer teacher-reported internalizing symptoms. These findings underscore the distinct role of pragmatic language in behavioural functioning.

Consistent with these findings, existing evidence suggests that less-developed pragmatic language abilities are associated with a broad range of socio-emotional difficulties, including increased depressive symptoms in late adolescence [51], general social-emotional problems [52], behavioural challenges [39], and elevated hyperactivity and attentional difficulties [53]. Together, these studies suggest that language abilities, and pragmatic language in particular, may represent a key developmental link between early language skills and later internalizing and externalizing symptoms, a hypothesis that has not been tested prospectively in comparison with core language skills.

The current study

The current study examined associations linking expressive vocabulary at age 2 to later internalizing and externalizing problems at age 8 using data from a large, prospective community-based cohort. Specifically, we investigated whether expressive vocabulary skills at age 2 were associated with internalizing and externalizing symptoms at age 8, and whether these associations are mediated by children’s core and pragmatic language skills at age 5.

Three research objectives guided the study. First, we examined whether expressive vocabulary at age 2 was associated with internalizing and externalizing symptoms in middle childhood. Second, we tested whether core language and pragmatic language at age 5 mediated associations between early expressive vocabulary and later internalizing and externalizing symptoms. Third, we compared the relative contributions of core and pragmatic language within these mediation models.

Based on prior literature, we hypothesized that expressive vocabulary at age 2 and core and pragmatic language abilities at age 5 would be associated with internalizing and externalizing symptoms at age 8. More central to our research questions, we hypothesized that both core and pragmatic language at age 5 would partially account for associations between expressive vocabulary at age 2 and internalizing and externalizing symptoms at age 8. Finally, given the proposed role of pragmatic language in social communication and behavioural functioning, we expected pragmatic language to demonstrate associations with internalizing and externalizing symptoms that were distinct from those observed for core language.

Methods

Participants

A total of 3,379 mother-child dyads from the All Our Families (AOF) dataset were included in these analyses. The AOF dataset is an ongoing longitudinal, community-based cohort based in Western Canada [54]. Recruitment of pregnant mothers occurred between 2008 and 2011 through community advertisements and local health facilities. Eligibility requirements included maternal age of at least 18 years, able to complete assessments in English, and less than 25 weeks’ gestation at the time of recruitment [54]. Informed consent was obtained from all participants. See Table 1 for demographic details of the sample. As is common in longitudinal cohort studies, attrition occurred due to factors such as aging out of assessments, funding delays, and COVID-19-related disruptions. For the current study, expressive vocabulary was assessed at age 2 (n = 1,594), core and pragmatic language at age 5 (n = 1,943) and internalizing (n = 1,820) and externalizing (n = 1,822) symptoms at age 8. Baseline covariates assessed at 24–27 weeks gestation were available for the full cohort (N = 3,379). Covariate-only cases (n = 1,008) were retained to maximize variability and preserve representativeness. The models were also estimated including only participants with data on at least one variable of interest (n = 2,379; expressive vocabulary, core language, pragmatic language, internalizing symptoms, or externalizing symptoms), and the pattern of results reported below remained unchanged.

Table 1.

All our families demographic characteristics

Variable	n (%)
Maternal education
Some elementary or high school	119 (3.5)
Graduated high school	248 (7.4)
Some college, trade, university	479 (14.3)
Graduated college, trade, university	1,979 (59.0)
Some graduate school	90 (2.7)
Completed graduate school	439 (13.1)
Total family income (in CAD)
Less than 50k	466 (14.2)
50k – 99.99k	1,151 (35.0)
100k or more	1,671 (50.8)
Child sex
Male	1,509 (52.8)
Female	1,348 (47.2)
Primary language
English	2,957 (88.1)
Non-English	401 (11.9)

Demographics calculated based on the baseline assessment administered during the mother’s pregnancy. CAD = Canadian Dollars. Sample size may vary across items as there were different responding patterns across participants (i.e., not every mother answered every question; therefore n does not always add up to 3,379)

Measures

Data were retrieved from the original baseline assessment completed around 25 weeks of maternal gestation, at 4 months postpartum, and when children were 2, 5, and 8 years of age. The baseline and 4-month postpartum assessments were included to obtain scores on the covariates of interest. The 2-, 5-, and 8-year time points were selected in order to allow for an investigation of the relations between language and internalizing and externalizing symptoms across developmental periods ranging from early to middle childhood.

Expressive vocabulary

Mothers completed the MacArthur-Bates Communicative Development Inventories: Words and Sentences [MBCDI-WS; 55] when their children were 2 years old. The parent-reported number of words produced by the child was used to obtain a measure of expressive vocabulary. Scores ranged from 0 to 680 and demonstrated excellent internal consistency (α = 0.95). Since the age range at this assessment varied from 22 to 34 months and the scores were not age-normed, child age at the time of assessment was included as a covariate.

Core and pragmatic language skills

Mothers completed the Children’s Communication Checklist-2 [CCC-2; 56] when children were 5 years old. The CCC-2 is a 70-item parent-report measure comprising 10 subscales assessing multiple aspects of children’s communication abilities. The CCC-2 manual provides a General Communication Composite and a Social Deviance Composite but does not provide standardized composite scores specifically representing core and pragmatic language domains [56]. However, Bishop described one purpose of the CCC as providing “a quantitative estimate of pragmatic language impairments in children" [57, 58, 59, p. 916]. Consistent with this conceptualization of pragmatic language as derived from CCC-2 subscales and distinguished from structural or core language abilities, prior research has frequently used CCC-2 subscales to derive a pragmatic language measure in children [e.g., 36, 49; see also 40 and 60 for a discussion of the use of the CCC-2 to measure pragmatic language].

In this study, scaled scores from four subscales assessing core language (speech, syntax, semantics, coherence) and four subscales assessing pragmatic language (initiation, scripted language, context, non-verbal communication) were summed to create researcher-derived composite scores. Both composites demonstrated good internal consistency (core: α = 0.86, pragmatic: α = 0.87). The remaining CCC-2 subscales (social relations and interests) were not included because they primarily index socio-behavioural characteristics and overlap conceptually with the behavioural outcome measures examined in the present study [56].

Prior factor analytic work on the CCC and CCC-2 has supported distinctions between structural/core language and pragmatic/social communication dimensions, although substantial overlap between constructs has consistently been reported [60]. To examine the appropriateness of our researcher-derived composite scores in our sample, we conducted a confirmatory factor analysis in R [61] using the lavaan package [64]. Model 1 included all eight CCC-2 subscales loading onto a single latent communication factor. Model 2 specified separate latent constructs representing core and pragmatic language, with 4 subscales loading onto each factor.

Model 2 demonstrated significantly improved fit compared to Model 1 (χ² = 21.54, df = 1, p < .001), supporting the distinction between core and pragmatic language dimensions within the present sample. The latent correlation between core and pragmatic language was high (r = .896), indicating substantive overlap. These findings are consistent with prior work suggesting that core language and pragmatic language as measured by the CCC-2, are closely related but distinguishable aspects of children’s communicative development [62–64].

Studies examining children’s language have primarily focused on domains such as vocabulary, grammar/syntax, and receptive versus expressive language, with pragmatic language rarely incorporated into these models [65–67]. The present CFA provides the potential to extend current knowledge on the dimensionality of language by demonstrating that pragmatic language can be meaningfully differentiated from core language using the CCC-2.

Internalizing and externalizing symptoms

Mothers completed the Parent Rating Scale (PRS) of the Behavior Assessment System for Children – Second Edition [BASC-2; 68] when children were 8 years old. The BASC-2 is a broad measure, comprised of 160 items assessing multiple aspects of child behaviour and functioning. Composite T scores for internalizing problems (anxiety, depression, somatization; 40 items) and externalizing problems (hyperactivity, aggression, conduct problems; 30 items) were included. Internal consistency was excellent for both scales (internalizing: α = 0.92; externalizing: α = 0.92).

Covariates

Several covariates were included based on prior literature demonstrating associations with language development and behavioural functioning, including maternal education, primary language spoken in the home, child sex, family income, and perinatal health risk factors [13, 20, 69–73]. Maternal education, income, and primary language were taken from the baseline assessment at approximately 25 weeks gestation. Maternal education was coded from 1 (some elementary or high school) to 6 (graduate school), with higher scores indicating higher levels of education. Family income was based on total income earned prior to taxes in the 12 months prior to baseline and was coded in CAD $10,000 increments, resulting in an 11-point scale. Primary language exposure indexed whether English was the primary language spoken in the home (1 = English, 0 = non-English).

Child sex and perinatal risk factors were retrieved from birth medical records. Perinatal risk was measured using the Antepartum Risk Score (APRS) which identifies risk for adverse pregnancy outcomes based on factors such as substance use, prior stillbirths, and gestational illness such as diabetes. APRS scores range from 0 to 16, with higher scores indicating a greater perinatal risk [74]. In the current study, the mean APRS score was 2.03 (SD = 2.11).

All covariates were initially included in preliminary models based on theoretical and empirical relevance. Because the APRS was not significantly associated with the primary study variables beyond expressive vocabulary, and inclusion of the APRS substantially increased model complexity without improving model fit, it was excluded from final structural equation models to facilitate model stability and interpretability. Importantly, retaining APRS in the models did not meaningfully alter the pattern of findings or interpretation of results.

Analytic strategy

To investigate associations between vocabulary skills at age 2 and internalizing and externalizing symptoms at age 8, with core and pragmatic language at age 5 as potential mediators, structural equation models were estimated in R [61] using the lavaan [75] and RMediation [76] packages. All models included expressive vocabulary at age 2 as the predictor and included covariates (maternal education, family income, child sex, and language spoken in the home) regressed onto the predictor, mediators, and outcome variables. Models 1, 1a, and 1b examined internalizing symptoms at age 8 as the outcome, mediated through core and pragmatic language at age 5. Models 2, 2a and 2b examined externalizing symptoms at age 8 as the outcome, mediated through core and pragmatic language at age 5. Prior to model estimation, all continuous variables were z-scored to create a consistent scale. Six final structural equation models were estimated and are reported in the results, three models examining internalizing symptoms at age 8 (one combined-mediator model and two single-mediator models) and three parallel models examining externalizing symptoms at age 8. Two preliminary versions of the combined mediator models (one examining internalizing symptoms and one examining externalizing symptoms) were also estimated with APRS retained as a covariate during model development.

All exogenous variables, including covariates and expressive vocabulary, were allowed to covary. Parameter estimates were calculated using 5,000 bootstraps per model. All available cases, including participants with covariate data only, were retained to maximize generalizability within our sample [77–79]. Therefore, the total sample size for all models was 3,379 children. Full information maximum likelihood (FIML) was used to handle missing data. Any missing data from the dependent, mediator, or outcome variables were consistent with missing completely at random (Little’s MCAR Test; χ² = 25.10, df = 18, p = .12).

To account for measurement error, single-indicator latent variables (SILVs) were constructed for the predictor, mediators, and outcomes using Cronbach’s alpha-based internal consistency estimates to estimate error variances [80]. All six models demonstrated acceptable to good fit. An important consideration when constructing these models was how much complexity should be added. The addition of SILVs increases consistency and reliability but also decreases model fit indices due to the latent variables being assigned fixed error variances, which constrain covariance structures [80]. When the models were estimated without SILVs, overall fit was excellent and the pattern of results remained unchanged.

Given that the primary research questions focused on potential indirect paths linking early expressive vocabulary, later language abilities, and behavioural outcomes, indirect effects were examined regardless of the significance of total associations between expressive vocabulary and later internalizing or externalizing symptoms. In support of this approach, contemporary mediation frameworks do not require a significant direct association between predictor and outcome variables for indirect effects to be estimated or interpreted [81, 82]. This is especially relevant in cases such as ours, where suppression or inconsistent mediation occurs. Since findings are not uniform across the field, it is possible that early language may not influence behaviour directly, but only through intermediary developmental processes.

Results

Descriptive statistics and correlations were calculated using jamovi 2.3.28 [83]. At the bivariate level, expressive vocabulary at age 2 was not significantly correlated with either internalizing or externalizing symptoms at age 8, whereas both core and pragmatic language at age 5 showed moderate associations with internalizing and externalizing symptoms at age 8. Expressive vocabulary was moderately correlated with both core language and pragmatic language, which were themselves highly correlated (see Table 2).

Table 2.

Bivariate Pearson’s correlations and descriptive statistics for the variables of interest

	n	M	SD	Range	1	2	3	4
1.Vocabulary skills	1,594	295	180	0-680
2.Pragmatic language	1,942	43	7.67	3.25–15.75	0.23***
3.Core language	1,943	44.5	8.09	1.5–15	0.41***	0.71***
4.Externalizing symptoms	1,822	49.6	8.29	34–96	− 0.04	− 0.36***	− 0.22***
5.Internalizing symptoms	1,820	49.1	10.10	30–107	0.05	− 0.29***	− 0.16***	0.46***

*** represents significance of p < .001

Model 1: expressive vocabulary x core and pragmatic language x internalizing behaviours

This model examined the associations among expressive vocabulary at age 2, core and pragmatic language at age 5, and internalizing symptoms at age 8, including indirect associations involving core language and pragmatic language (see Fig. 1). The model fit was good (RMSEA = 0.06, 90% CI [0.06, 0.07]; CFI = 0.95, TLI = 0.82; SRMR = 0.07). Expressive vocabulary at age 2 was positively associated with both core language (B = 0.35, SE = 0.02, p < .001, β = 0.32) and pragmatic language (B = 0.17, SE = 0.03, p < .001, β = 0.17) at age 5. Pragmatic language showed a significant negative association with internalizing symptoms at age 8 (B = − 0.44, SE = 0.06, p < .001, β = − 0.43) in the model accounting for early vocabulary, core language, and covariates. In contrast, core language showed a positive association with internalizing behaviours (B = 0.13, SE = 0.06, p = .037, β = 0.13). Finally, a small residual direct path was found from vocabulary skills to internalizing symptoms (B = 0.07, SE = 0.03, p = .016, β = 0.07). Significant indirect associations were observed between early expressive vocabulary and later internalizing symptoms through both pragmatic language (B = − 0.08, SE = 0.02, p < .001, β = − 0.07) and core language (B = 0.04, SE = 0.02, p = .040, β = 0.04).

Fig. 1.

Model 1: Expressive Vocabulary x Core and Pragmatic Language x Internalizing Behaviours. *** represents significance of p < .001, * represents significance of p < .05. Unstandardized path coefficients are shown in the figure

Core and pragmatic language were highly correlated (r = .71, p < .001), indicating substantial shared variance between constructs. In the combined mediator model, associations between language variables and internalizing symptoms reflected conditional associations estimated after accounting for overlap between core and pragmatic language. This resulted in evidence of inconsistent indirect effects, wherein associations involving pragmatic language (negative effect) and core language (positive effect) differed in direction within the combined model [84]. Consequently, findings from the combined model should be interpreted cautiously and may not reflect the associations between core language and internalizing symptoms reported in prior studies [18]. To further examine these relations, simple mediation models were estimated separately for core and pragmatic language using identical predictors, outcomes, and covariates as those in Model 1.

Model 1a: core language only

Model 1a examined the associations among expressive vocabulary at age 2, core language at age 5, and internalizing symptoms at age 8, including indirect associations involving core language (see Fig. 2). The model fit was acceptable (RMSEA = 0.06, 90% CI [0.06, 0.07]; CFI = 0.89, TLI = 0.67; SRMR = 0.06). Vocabulary skills were positively associated with core language skills (B = 0.35, SE = 0.02, p < .001, β = 0.32). The association between core language and internalizing symptoms differed from that observed in the combined model above, becoming negative in the simplified model (B = − 0.23, SE = 0.03, p < .001, β = − 0.24). Given the substantial overlap between core and pragmatic language, this pattern is consistent with a suppressor effect in the combined model. Vocabulary skills had a positive residual association with internalizing symptoms after accounting for core language (B = 0.11, SE = 0.03, p < .001, β = 0.11). Significant indirect associations were observed between early vocabulary and later internalizing symptoms through core language (B = − 0.08, SE = 0.01, p < .001, β = − 0.08). In contrast to the combined model, a larger expressive vocabulary was associated with stronger core language, which in turn was associated with fewer internalizing symptoms when pragmatic language was not simultaneously included in the model.

Fig. 2.

Model 1a: Expressive Vocabulary x Core Language x Internalizing Behaviours. *** represents significance of p < .001. Unstandardized path coefficients are shown in the figure

Model 1b: pragmatic language only

Model 1b examined associations among expressive vocabulary at age 2, pragmatic language at age 5, and internalizing symptoms at age 8 (see Fig. 3). The model fit was good (RMSEA = 0.05, 90% CI [0.04, 0.06]; CFI = 0.93, TLI = 0.78; SRMR = 0.05). Expressive vocabulary was positively associated with pragmatic language skills (B = 0.20, SE = 0.03, p < .001, β = 0.19), and pragmatic language was negatively associated with internalizing symptoms at age 8 (B = − 0.33, SE = 0.03, p < .001, β = − 0.33). A positive residual association was observed between expressive vocabulary and internalizing symptoms after accounting for pragmatic language (B = 0.09, SE = 0.03, p = .001, β = 0.09). Significant indirect associations were observed between early expressive vocabulary and later internalizing symptoms through pragmatic language (B = − 0.07, SE = 0.01, p < .001, β = − 0.06). That is, a larger expressive vocabulary was associated with stronger pragmatic language, which in turn was associated with fewer internalizing symptoms in the single-mediator model.

Fig. 3.

Model 1b: Expressive Vocabulary x Pragmatic Language x Internalizing Behaviours. *** represents significance of p < .001, ** represents significant of p < .01. Unstandardized path coefficients are shown in the figure

Model 2: expressive vocabulary x core and pragmatic language x externalizing behaviours

This model examined the associations among expressive vocabulary at age 2, core and pragmatic language at age 5, and externalizing symptoms at age 8, including indirect associations involving core and pragmatic language (see Fig. 4). The model fit was good (RMSEA = 0.07, 90% CI [0.06, 0.08]; CFI = 0.94, TLI = 0.80; SRMR = 0.07). Expressive vocabulary at age 2 was positively associated with both core language (B = 0.34, SE = 0.02, p < .001, β = 0.32) and pragmatic language (B = 0.17, SE = 0.03, p < .001, β = 0.17) at age 5. Pragmatic language skills showed a significant negative association with externalizing symptoms at age 8 (B = − 0.51, SE = 0.06, p < .001, β = − 0.49) in the model accounting for early vocabulary, core language, and covariates. In contrast, core language skills showed a positive association with externalizing symptoms (B = 0.18, SE = 0.05, p = .001, β = 0.18). Again, given the substantial overlap between core and pragmatic language, this pattern was consistent with a suppressor effect in the combined model. In contrast to the internalizing models, no significant residual direct association was observed between expressive vocabulary and externalizing symptoms (B = 0.03, SE = 0.03, p = .34, β = 0.03). Significant indirect associations were observed between early expressive vocabulary and later externalizing symptoms through both pragmatic language (B = − 0.09, SE = 0.02, p < .001, β = − 0.08) and core language (B = 0.06, SE = 0.02, p = .001, β = 0.06).

Fig. 4.

Model 2: Expressive Vocabulary x Core and Pragmatic Language x Externalizing Behaviours. *** represents significance of p < .001, ** represents significance of p < .01. Unstandardized path coefficients are shown in the figure

As in the internalizing models, core and pragmatic language were highly correlated, and the combined model produced opposing conditional associations with externalizing symptoms, consistent with inconsistent mediation arising from shared variance between the language domains. To further examine these relations, follow-up single-mediator models were estimated separately for core and pragmatic language using identical predictors, outcomes, and covariates.

Model 2a: core language only

Model 2a examined associations among expressive vocabulary at age 2, core language at age 5, and externalizing symptoms at age 8, including indirect associations involving core language (see Fig. 5). The model fit was acceptable (RMSEA = 0.07, 90% CI [0.06, 0.08]; CFI = 0.89, TLI = 0.67; SRMR = 0.06). Expressive vocabulary was significantly and positively associated with core language skills (B = 0.35, SE = 0.02, p < .001, β = 0.32). When examined without simultaneously accounting for pragmatic language, core language was negatively associated with externalizing symptoms (B = − 0.23, SE = 0.03, p < .001, β = − 0.24). A positive residual association was observed between expressive vocabulary and externalizing symptoms (B = 0.08, SE = 0.03, p = .011, β = 0.07). Significant indirect associations were observed between early expressive vocabulary and later externalizing symptoms through core language (B = − 0.08, SE = 0.01, p < .001, β = − 0.08). Thus, a larger expressive vocabulary was associated with stronger core language, which in turn was associated with fewer externalizing symptoms in the single-mediator model.

Fig. 5.

Model 2a: Expressive Vocabulary x Core Language x Externalizing Behaviours. *** represents significance of p < .001, * represents significance of p < .05. Unstandardized path coefficients are shown in the figure

Model 2b: pragmatic language only

Model 2b examined associations among expressive vocabulary at age 2, pragmatic language at age 5, and externalizing symptoms at age 8, including indirect associations through pragmatic language (see Fig. 6). The model fit was good (RMSEA = 0.06, 90% CI [0.05, 0.07]; CFI = 0.92, TLI = 0.74; SRMR = 0.05). Expressive vocabulary was positively associated with pragmatic language (B = 0.20, SE = 0.03, p < .001, β = 0.19), and pragmatic language was negatively associated with externalizing symptoms (B = − 0.37, SE = 0.03, p < .001, β = − 0.36). A significant positive residual association was observed between expressive vocabulary and externalizing symptoms after accounting for pragmatic language (B = 0.06, SE = 0.03, p = .016, β = 0.06). Significant indirect associations were observed between early expressive vocabulary and later externalizing symptoms through pragmatic language (B = − 0.07, SE = 0.01, p < .001, β = − 0.07). Thus, a larger expressive vocabulary was associated with stronger pragmatic language, which in turn was associated with fewer externalizing symptoms in the single-mediator model.

Fig. 6.

Model 2b: Expressive Vocabulary x Pragmatic Language x Externalizing Behaviours. *** represents significance of p < .001, * represents significance of p < .05. Unstandardized path coefficients are shown in the figure

Comparisons between models

Differences in indirect associations across the internalizing and externalizing models were examined using z-score analyses. Indirect effects were comparable in magnitude across the internalizing and externalizing models, both in the combined models and in the single-mediator models for core and pragmatic language (all |z| < 0.62, ps > 0.54), suggesting similar overall patterns of indirect associations across behavioural domains.

Given evidence of substantial overlap between core and pragmatic language, as well as inconsistent indirect associations in the combined models, comparisons between core and pragmatic language were limited to the single-mediator models. Associations between language skills at age 5 and later behavioural symptoms differed between core and pragmatic language. Pragmatic language showed larger conditional associations with both internalizing (z = -2.70, p = .007) and externalizing symptoms (z = -3.45, p < .001) relative to core language after accounting for early expressive vocabulary and covariates. However, indirect associations were comparable in magnitude across the core and pragmatic language models.

Discussion

The current study examined longitudinal associations among expressive vocabulary at age 2, core and pragmatic language at age 5, and internalizing and externalizing symptoms at age 8. Three primary findings emerged. First, early expressive vocabulary was not significantly associated with later behavioural symptoms at the bivariate level and showed only small and inconsistent residual associations within structural equation models. Second, indirect associations linking early expressive vocabulary and later internalizing and externalizing behaviour were observed through both core language and pragmatic language. Third, indirect associations were comparable in magnitude across core and pragmatic language models, suggesting that multiple components of language development may relate to later behavioural functioning.

Early expressive vocabulary, later language, and behavioural outcomes

Expressive vocabulary at age 2 was not significantly associated with either internalizing or externalizing behaviours at age 8 at the bivariate level. Expressive vocabulary was positively associated with both core and pragmatic language at age 5, and both language scores were subsequently associated with later behavioural symptoms. It is possible that expressive vocabulary measured at age 2 represents an earlier and less stable stage of language development than the broader constructs assessed at age 5, which may partially explain the relatively modest longitudinal associations observed across developmental periods [85].

Indirect associations between early vocabulary and later internalizing and externalizing symptoms were observed and were comparable in magnitude for both core and pragmatic language, suggesting that structural and social aspects of language may help explain how early vocabulary relates to later behavioural functioning. These findings align with prior research demonstrating longitudinal associations between early language abilities and later socioemotional development [10–14]. The present findings extend this literature by examining both core and pragmatic language within a large, community-based longitudinal sample.

Comparison of indirect associations through core and pragmatic language

The present findings suggest that early expressive vocabulary was indirectly associated with later behavioural symptoms through both core and pragmatic language. These results extend the findings from previous research demonstrating longitudinal associations between language abilities and later internalizing and externalizing symptoms by examining both core and pragmatic language within a large, community-based longitudinal cohort [13, 18, 26, 28].

Interpretation of these findings requires consideration of the substantial overlap between core and pragmatic language. Recall that the confirmatory factor analysis supported a distinction between core and pragmatic language while also demonstrating substantial overlap between constructs. These findings suggest that core and pragmatic language represent highly related but potentially distinguishable aspects of communication that may each relate to later behavioural functioning. When both constructs were included simultaneously in the combined models, pragmatic language was negatively associated with both internalizing and externalizing behaviours, whereas core language was positively associated. Follow-up single-mediator models indicated that stronger core language was associated with fewer internalizing and externalizing symptoms when pragmatic language was not simultaneously included in the models. This pattern is consistent with suppressor effects that can emerge when highly correlated predictors are entered simultaneously into mediation models [80, 86, 87]. Such effects arise from the partitioning of shared variance and do not necessarily reflect distinct developmental processes. Therefore, findings from the combined models should be interpreted cautiously, as the estimated associations reflect conditional associations after accounting for shared variance between the language constructs. Instead, we focus our discussion on results from the single mediator models.

The present findings suggest that pragmatic language skills may represent one aspect of children’s broader communication abilities associated with later behavioural functioning. Importantly, these associations were observed within a community sample, consistent with results from clinical populations [42, 44, 47], suggesting that links between social communication and behavioural functioning may extend across a broader and more diverse population of children than previously demonstrated. From a developmental perspective, pragmatic language may relate to behavioural adjustment because it reflects children’s ability to use language effectively in social interaction, emotion understanding, and behavioural regulation [36, 88]. Pragmatic skills involve interpreting nonverbal cues, maintaining conversational coherence, repairing misunderstandings, and adapting language to social contexts [24, 34]. These abilities are closely linked to successful peer interactions and emotional functioning [37, 88]. Prior studies have shown that children with stronger emotion recognition and social communication skills exhibit fewer internalizing and externalizing problems [89–91]. The present findings extend this literature by demonstrating that pragmatic language is associated with behavioural outcomes even after accounting for early vocabulary and core language abilities, although these associations should be interpreted in the context of the overlap between language constructs.

In parallel, the present findings indicate that core language abilities may relate to behavioural adjustment [20, 28, 31, 32], perhaps through somewhat different developmental processes. Structural language skills such as vocabulary, syntax, semantics, and coherence support children’s ability to understand spoken language, organize thoughts, communicate needs clearly, and participate in academic and social environments [92]. Difficulties in these areas may contribute to frustration, misunderstandings, emotional dysregulation, and reduced participation in peer and classroom interactions [12, 18, 28].

Limitations and future directions

The present study had several strengths, including the use of well-validated measures of child expressive vocabulary and core and pragmatic language [55, 56] and measures of internalizing and externalizing symptoms [68] across a large, prospective longitudinal cohort. The findings, however, should be considered in light of several limitations. First, although the sample size is large and largely representative of the geographic area where the study took place, the sample was primarily composed of families with middle to upper socio-economic status, which may limit generalizability of the findings [54]. Second, the language and behavioural measures were based on maternal reports. Although parent-report measures such as the CCC-2 and BASC-2 are widely used and have excellent psychometric properties, they do not capture all aspects of child language and behavioural functioning across settings and may therefore provide a narrower lens than what could be captured across use of multiple different measure formats. Third, the reliance on maternal report increases the possibility of shared method variance and inflated associations among language and behavioural constructs.

Fourth, the current study focused primarily on expressive aspects of language development. Future work should investigate whether other early communicative skills, including receptive language, gesture use, and early social communication, show similar indirect associations with later behavioural functioning. Additionally, pragmatic language could be explored further by including more specific social-communication measures, such as conversational repair, peer interaction quality, and perspective taking. Although detailed observational and interaction-based assessments would be difficult to implement within large longitudinal cohorts, future research could incorporate targeted pragmatic questionnaires, teacher-report measures, or smaller experimental and clinical studies examining specific pragmatic skills in greater depth. These broader assessments may help to further clarify the overlap and distinctions between structural and pragmatic aspects of language development. Furthermore, incorporating multiple informants, observational measures of communication, and broader assessments of social-cognitive functioning may also help further disentangle overlapping aspects of language development and behavioural adjustment.

Fifth, due to constraints associated with the cohort design and the need to balance model complexity and interpretability, only maternal education, primary language spoken in the home, child sex, and family income were included as covariates. Although APRS was initially included based on theoretical relevance, it was removed from final models due to minimal associations with the primary study variables and increased model complexity. This model-trimming approach may influence model stability and reproducibility. In addition, language development and behavioural functioning emerge within broader developmental systems impacted by parenting quality, child cognitive ability, home language environment, and early behavioural problems [93, 94] that were not measured in the present study.

Finally, the current study tested a unidirectional model from early vocabulary to later behavioural outcomes. However, some research suggests that relations between language development and behavioural functioning may be bidirectional, with early behavioural difficulties also influencing subsequent language growth [19]. Furthermore, the models run in this study test associative relations and cannot directly be tied to developmental mechanisms. Despite these limitations, the present study highlights the importance of considering multiple language domains, across different time points when examining their link to later behavioural adjustment.

Future research should continue to examine longitudinal associations between language development and behavioural functioning across later developmental periods, including adolescence. Adolescence represents an important period for the emergence of internalizing and externalizing difficulties [95], while also involving substantial social, emotional, cognitive, and interpersonal developmental changes that may alter associations between language and internalizing and externalizing symptoms [96]. Extended longitudinal follow-up may help clarify how associations involving core language, pragmatic language, and behavioural functioning evolve across development and whether different developmental periods represent heightened periods of vulnerability or adaptation.

Implications

The findings may have implications for understanding how different aspects of language development relate to later behavioural functioning. Although the current study cannot establish causal mechanisms or directly inform intervention effectiveness, the findings, along with prior research, suggest that multiple aspects of language development, including both core and pragmatic language, are associated with later behavioural functioning. The findings suggest that core and pragmatic language are highly related, but distinguishable, aspects of language that both interact with later behaviour.

These findings support prior literature suggesting that broader assessment of children’s communication abilities, beyond vocabulary skills alone, may be important when considering children’s behavioural functioning. Prior research indicates that language interventions can have a positive impact on children’s behavioural outcomes, particularly when interventions target functional communication and social interaction [97]. Supporting social-communication skills such as turn-taking, conversational coherence, and interpretation of social cues may be especially relevant for promoting behavioural adjustment [98, 99]. Although much of the intervention literature targeting pragmatic language has focused on clinical populations (e.g., children with developmental language disorder or autism), the present findings suggest that associations between social communication and behavioural functioning may also extend to community-based samples. Previous intervention literature suggests that socially embedded aspects of communication may serve as a viable and developmentally appropriate target for early support within community settings [100, 101].

Conclusions

This study found indirect associations linking early expressive vocabulary and later behavioural symptoms through both core and pragmatic language, highlighting the importance of considering both structural and social aspects of communication when examining children’s behavioural adjustment. These findings contribute to a growing literature highlighting associations between language development and socioemotional functioning across childhood.

Author contributions

AB: Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing. LR: Conceptualization, Writing – original draft, Writing – review & editing. KW: Software, Writing – review & editing. ST: Funding acquisition, Resources, Project administration, Data curation, Writing – review & editing. SG: Conceptualization, Methodology, Writing – original draft, Writing – review & editing, Supervision, Funding acquisition. All authors read and approved the final manuscript. AB and LR share first authorship and ST and SG share senior authorship.

Funding

The All Our Families study is supported by funding from Alberta Innovates Interdisciplinary Team Grant 200700595, the Alberta Children’s Hospital Foundation, the Max Bell Foundation, and additional grants awarded to Suzanne C. Tough. Lorraine Reggin’s contributions were supported through her Researcher in Residence position at the Owerko Centre, funded by the Westman Foundation. Karolina Wieczorek was supported by a graduate scholarship from the Social Sciences and Humanities Research Council (SSHRC). Susan A. Graham’s contributions to this research were funded by the Alberta Children’s Hospital Foundation through the Owerko Centre and a SSHRC Insight Grant.

Data availability

The All Our Families Cohort data are available through requests to the study (https://ucalgary.ca/allourfamilies), and metadata of variables can be found at https://www.maelstrom-research.org/study/aof. Restrictions apply to the availability of these deidentified data, which were used according to data sharing agreements.

Declarations

Ethics approval and consent to participate

Ethics approval to use the All Our Families data for this study was granted by the University of Calgary Conjoint Health Research Ethics Board (REB21-0428).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.