Language and Inhibition: Predictive Relationships in Children With Language Impairment Relative to Typically Developing Peers

Abstract

Background

This study examined predictive relationships between two indices of language—receptive vocabulary and morphological comprehension—and inhibition in children with specific language impairment (SLI) and typically developing (TD) children.

Methods

Participants included 30 children with SLI and 41 TD age-matched peers (8–12 years). At two time points separated by 1 year, we assessed receptive vocabulary and morphological comprehension via standardized language measures and inhibition via a Flanker task. We used Bayesian model averaging and Bayesian regression analytical techniques.

Results

Findings indicated predictive relationships between language indices and inhibition reaction time (RT), but not between language indices and inhibition accuracy. For the SLI group, Year 1 inhibition RT predicted Year 2 morphological comprehension. For the TD group, Year 1 morphological comprehension predicted Year 2 inhibition RT.

Conclusions

This study provides preliminary evidence of a predictive relationship between language and inhibition, but this relationship differed between children with SLI and those with typical development. Findings suggest that inhibition RT played a larger predictive role in later morphological comprehension in children with SLI relative to the other relationships examined. Targeting inhibition skills as a part of language intervention may improve subsequent morphological comprehension.

Supplemental Material

https://doi.org/10.23641/asha.12014823

Children with specific language impairment (SLI) have significant deficits in language relative to typical peers in the absence of hearing impairment, medical or developmental disability, or clinical deficits in nonverbal IQ (NVIQ), with an estimated prevalence of 7% (Norbury et al., 2016; Tomblin et al., 1997; see also Bishop et al., 2017, for an alternative diagnostic label—developmental language disorder). In addition to linguistic deficits, children with SLI have deficits outside the realm of language, such as in executive function (EF) skills (Ebert & Kohnert, 2011; Vugs et al., 2015). EF is a key set of cognitive processes—working memory (WM), inhibition, cognitive flexibility, and a central executive attentional system—involved in goal-directed behavior (Marcovitch & Zelazo, 2009; Miyake et al., 2000). There is a substantive literature linking EF with long-term academic and vocational outcomes (Best et al., 2011; Moffitt et al., 2011), such as inhibition and academic ability (van der Schoot et al., 2000).

Inhibition involves resisting irrelevant information, such as distractors (also referred to as “interference control”; Kaushanskaya et al., 2017; Pauls & Archibald, 2016; Spaulding, 2010), and it is implicated in incremental linguistic processing (Ibbotson & Kearvell-White, 2015; Marton et al., 2018; Miller et al., 2001). Children with SLI have reliable deficits in inhibition (Ebert & Kohnert, 2011; Pauls & Archibald, 2016), and the relationship between language ability and inhibition may be different in SLI relative to typical development (Dispaldro et al., 2013). These group differences, however, are not well defined, nor is the directionality of influence between language and inhibition.

Relationship Between Language and Inhibition

There are several hypotheses of the directionality of influence between language and inhibition. The hierarchical competing systems model (HCSM) links language and EF through the role of language-based reflection in over-riding prepotent responses during goal-directed behavior (Marcovitch & Zelazo, 2009). Under this model, learners who use language to reflect on behavior will have more optimal EF performance than learners who do not use language to reflect on behavior. Inhibitory functions draw on language-based reflection in situations requiring inhibition of a previously correct response in favor of a new response (Marcovitch & Zelazo, 2009), such as abandoning an early misinterpretation of a sentence for a subsequent correct interpretation (e.g., late-resolving ambiguity; Minai et al., 2012).

Links between language and inhibition in SLI may relate to hypotheses of processing-based causal factors in SLI. The generalized slowing hypothesis (Kail, 1994; Miller et al., 2001) is based on slowed reaction times (RTs) in SLI relative to typically developing (TD) peers in verbal and nonverbal tasks. Language processing occurs incrementally over multiple time scales, and inefficient coordination of activation and suppression is associated with poorer rather than better language skills (Marton et al., 2014; Miller et al., 2001). In fact, the inefficient inhibition hypothesis suggests that children with SLI may have difficulty with language due to slower and less accurate inhibitory functions, which allow irrelevant information to enter WM and draw WM processing away from relevant information (Bjorklund & Harnishfeger, 1990; Marton et al., 2007). While the HCSM suggests that language influences inhibition and other EF skills in TD children, the inefficient inhibition hypothesis suggests that inhibition influences language skills in children with SLI. No prior work, however, has tested the relationship between language and inhibition over time in children with SLI in order to resolve these conflicting theories.

The relationship between language and inhibition may differ according to specific language indices. Although time scales for processing words and morphemes may be similar, the long-range dependencies associated with morpheme comprehension (e.g., inflectional morphemes) likely draw on inhibitory functions to a greater degree than the shorter range dependencies associated with vocabulary comprehension (e.g., syllable identification; Dispaldro et al., 2013; Ibbotson & Kearvell-White, 2015). Indeed, school-age children with SLI, who have deficits in inhibition relative to TD peers, often have less severe impairments in vocabulary than morphology (Dispaldro et al., 2013; Leonard, 2014; Pauls & Archibald, 2016; see the procedural deficit hypothesis, Ullman & Pierpont, 2005, and work on statistical learning, Hsu & Bishop, 2011; Obeid et al., 2016).

Empirical work further motivates examining specific connections between language and inhibition in school-age children. In TD children, Kaushanskaya et al. (2017) found associations between a syntax composite, but not a semantics composite, and concurrent performance on a nonverbal inhibition measure controlling for age, socioeconomic status (SES), and NVIQ. In children with SLI, vocabulary (Finneran et al., 2009; Marton et al., 2014) and composite language (Marton et al., 2018) have been linked to concurrent inhibition performance, yet some work has found no significant association between language and inhibition performance (Bishop & Norbury, 2005). Dispaldro et al. (2013) showed a significant concurrent association between syntactic, but not semantic, comprehension and performance on a nonverbal inhibition task in children with SLI (Dispaldro et al., 2013). These conflicting results highlight the need to further characterize the relationship between language and inhibition in children with SLI.

Inhibition in SLI

Considerable evidence demonstrates inhibition deficits in SLI regardless of whether the task is verbal or nonverbal (Ebert & Kohnert, 2011; Pauls & Archibald, 2016; Spaulding, 2010; Victorino & Schwartz, 2015). The degree of deficit appears to be related to aspects of task demands (e.g., abstract vs. familiar stimuli; Roebuck et al., 2018) and group characteristics (e.g., relative severity of syntactic deficits; Dispaldro et al., 2013; Roebuck et al., 2018).

Individual Differences in Language and Inhibition in SLI

Individual differences in inhibition performance have been linked to language ability in TD children (Archibald et al., 2015; Kaushanskaya et al., 2017) and children with SLI (Finneran et al., 2009; Ibbotson & Kearvell-White, 2015; Marton et al., 2014), although the relationship between language and inhibition may differ between SLI and TD groups (Dispaldro et al., 2013). Inhibition accuracy has been found to share a significant concurrent correlation with receptive grammar in an SLI group, but not a TD group (Dispaldro et al., 2013), and with receptive (Finneran et al., 2009) and expressive vocabulary when combining SLI and TD groups (Marton et al., 2014). Other studies, however, found only weak correlations between inhibition and language when analyzing TD and SLI groups combined (Bishop & Norbury, 2005; Ladányi & Lukács, 2016). Critically, only Dispaldro et al. (2013) analyzed SLI versus TD group differences in the relationship between language and inhibition performance. Analyzing SLI and TD group differences in the relationship between language and inhibition using separate indices of language may reveal novel insights and clarify discrepancies in the literature (Dispaldro et al., 2013; Kaushanskaya et al., 2017). Additionally, no prior work has examined these relationships over time in order to characterize directionality of influence. The question of causality has been raised in prior literature, yet there are few data that address causal relationships between language and inhibition (Finneran et al., 2009; Kaushanskaya et al., 2017). A better understanding of the directionality of influence between language and inhibition will inform theoretical models and may further elucidate processing-based mechanisms underlying deficits in children with SLI.

The Current Study

The current study examined the association between two indices of language and inhibition performance of children with SLI relative to TD peers. We minimized verbal demands of the inhibition task in order to reduce the confound of having a substantive linguistic component in our measure of inhibition. Our language measures included an assessment of morphological comprehension (grammaticality judgment) and receptive vocabulary (nouns, verbs, adjectives). We also examined performance over time in order to explore directionality of influence between these constructs.

The aims of this project were to

1. examine predictive relationships between Year 1 language indices, namely, morphological comprehension and receptive vocabulary, and Year 2 inhibition performance, namely, RT and accuracy, in children with SLI relative to TD peers; and

2. examine predictive relationships between Year 1 inhibition performance, namely, RT and accuracy, and Year 2 language indices, namely, morphological comprehension and receptive vocabulary, in children with SLI relative to TD peers.

We hypothesized that morphological comprehension and inhibition performance would be more closely related than receptive vocabulary and inhibition performance based on empirical evidence from prior work (Dispaldro et al., 2013; Ibbotson & Kearvell-White, 2015). We also hypothesized that relationships between language and inhibition would differ between groups based on evidence from prior work (Dispaldro et al., 2013) and other findings from our lab (Ellis Weismer et al., 2017; Larson et al., 2019).

Method

The Education and Social Behavioral Institutional Review Board at the University of Wisconsin-Madison approved this study. Informed consent was obtained from parents in written form, and children provided verbal assent. Participants were recruited from the Madison metropolitan region and neighboring school districts as a part of a larger study on EF in four child populations at two time points approximately one year apart. At Year 1, the current study included 35 children with SLI (M = 9.7 years, SD = 1.2 years) and 48 age-matched TD children (M = 9.5 years, SD = 1.0 years; p = .65), and at Year 2, 30 children with SLI (M = 10.8 years, SD = 1.1 years) and 41 TD children (M = 10.6 years, SD = 1.0 years; p = .42) remained.

Eligibility for the study required passing a 1000-, 2000-, and 4000-Hz hearing screening at 20 dB, mono-lingualism, normal or corrected-to-normal vision, and no known history of autism, psychiatric, or neurological disorders per parental report. The TD group had no history of language delay or intervention based on parental report and standard scores ≥ 85 for NVIQ on the Wechsler Intelligence Scale for Children–Fourth Edition (WISC-IV; Wechsler, 2003) and for Expressive and Receptive Language on the Clinical Evaluation of Language Fundamentals–Fourth Edition (CELF-4; Semel et al., 2003).

The SLI group was recruited according to traditional criteria: NVIQ standard scores ≥ 85 (WISC-IV) and CELF-4 Expressive or Receptive standard scores ≤ 1.25 SDs below the mean, in addition to overall CELF-4 Core Language ≤ 1 SD below the mean. Total years of maternal education served as a proxy for SES (1 = first grade, 12 = high school, > 12 = postsecondary). There were significant group differences for SES (p ˂ .05) and NVIQ (p ˂ .001) at Year 1, but groups were similar on gender (SLI: 57% male; TD: 56% male; p = .96). Participant race demographics were 65% White, 22% African American, 1% Native American, 1% Asian, and 11% other, and 7% of participants were of Hispanic/Latino ethnicity per parental report.

We used standard scores from the Test of Language Development: Intermediate–Fourth Edition Morphological Comprehension subscale (Morphology; reliability coefficient = .97; Hammill & Newcomer, 2008) and the Peabody Picture Vocabulary Test–Fourth Edition (Vocabulary; reliability coefficient = .96–.97; Dunn & Dunn, 2007) for our language variables. The Morphology assessment involves a dichotomous grammaticality judgment (e.g., noun–verb agreement), and the Vocabulary assessment involves identifying a picture from a field of four options associated with a spoken word. Baseline visual RT and baseline auditory RT measures were also administered. Baseline visual RT was included in statistical models as it was significantly different between groups at Year 1 (p = .036); this difference was not significant at Year 2 (p = .085; Willoughby et al., 2018). We did not include baseline auditory RT in statistical models as groups did not differ at Year 1 (p = .21) or Year 2 (p = .60). Parents (typically mothers) completed the Conners Rating Scales attention-deficit/hyperactivity disorder (ADHD) index (Conners et al., 2008) at Year 1, and t scores from this measure were included in statistical models due to evidence of group differences (p < .001). NVIQ (WISC-IV) standard scores were also included in statistical models due to evidence of group differences (p < .001; see Table 1).

Table 1.

Demographic information and performance on standardized tests.

Participant characteristics	TD Year 1 (n = 48)	TD Year 2 (n = 41)	SLI Year 1 (n = 35)	SLI Year 2 (n = 30)	Year 1, p value	Year 2, p value
Age in years
M	9.54	10.60	9.65	10.80	p = .65	p = .424
SD	1.00	0.97	1.16	1.14	(r 2 = −.01)	(r 2 = −.01)
Maternal education in years (SES)
M	17.07	NA	15.17	NA	TD > SLI*	NA
SD	3.1		4.15		(r 2 = .05)
WISC-IV (NVIQ)
M	109.56	NA	96.20	NA	TD > SLI***	NA
SD	11.97		11.30		(r 2 = .24)
Conners ADHD Index
M	61.29	NA	75.92	NA	TD > SLI***	NA
SD	17.01		16.00		(r 2 = .15)
Core Language (CELF-4)
M	107.94	NA	78.09	NA	TD > SLI***	NA
SD	12.36			7.91	(r 2 = .65)
Receptive Vocabulary (PPVT-4)
M	117.02	116.08	96.06	95.90	TD > SLI***	TD > SLI***
SD	17.88	16.49	10.73	11.89	(r 2 = .31)	(r 2 = .31)
Morphological Comprehension (TOLD-4)
M	11.09	10.90	7.70	7.35	TD > SLI***	TD > SLI***
SD	2.69	3.27	2.02	2.08	(r 2 = .30)	(r 2 = .27)

Note. TD = typically developing; SLI = specific language impairment; r ² = adjusted r ²; SES = socioeconomic status; NA = not applicable; WISC-IV = Wechsler Intelligence Scale for Children–Fourth Edition; Conners ADHD Index = Conners Attention-Deficit/Hyperactivity Disorder Index–Parent Report, t scores; CELF-4 = Clinical Evaluation of Language Fundamentals–Fourth Edition; PPVT = Peabody Picture Vocabulary Test–Fourth Edition; TOLD-4 = Test of Language Development: Intermediate–Fourth Edition. Standard scores are reported for the WISC-IV, CELF-4, PPVT, and TOLD-4.

^*Group difference at p < .05.

^***Group difference at p < .001.

Experimental Task

Flanker Task

The Flanker task is a widely used measure of inhibition (Mullane et al., 2009; Salthouse, 2010) that involves resisting simultaneous interference. Our version of the task minimized verbal demands by using visual target and interference stimuli (images of fish and seaweed) and verbal instructions that were made simple by pairing them with corresponding visual representations and gestures (e.g., “If the fish is swimming in this direction [point to fish and show direction with finger], press this button [point to right button]”). The child was instructed to push buttons on a response box indicating the direction the center target stimulus was facing while ignoring surrounding interference stimuli. Practice trials involved six untimed trials with feedback followed by six timed trials with feedback to ensure that all participants understood the task. Research assistants used visual feedback as much as possible to further minimize verbal task demands. Participants completed 72 timed trials across three conditions (42 trials each): neutral (center fish surrounded by seaweed), congruent (fish facing the same direction; left or right), and incongruent (center fish surrounded by fish facing the opposite direction; left or right). Our outcome measures were accuracy and RT for the incongruent condition. RT analyses included accurate responses only, and we removed outliers ≥ 2.5 SDs from the mean RT for each participant (2.53% of the data). It should be noted that we included baseline visual RT measures in statistical analyses in order to model baseline RT as a predictor of performance beyond RT in the Flanker incongruent condition (Willoughby et al., 2018).

A prior study from our lab conducted a principal component analysis on EF tasks, including the Flanker task's incongruent and congruent conditions, in TD school-age children (Kaushanskaya et al., 2017). This analysis revealed that performance in the Flanker incongruent condition, but not the Flanker congruent condition, loaded onto the inhibition factor (see also Ellis Weismer et al., 2017; Engel De Abreu et al., 2014). This finding suggests that the incongruent condition has greater specificity in measuring inhibition. Alternatively, performance in the congruent condition may represent a broader, generalized ability to engage in the Flanker task, potentially capturing lower level EF skills such as sustained attention (Garon et al., 2008). Indeed, our analyses of the congruent condition align with this interpretation. Relationships between performance in the congruent condition and language indices were present in the context of several, if not all, participant characteristics (e.g., SES, NVIQ, baseline visual RT, ADHD symptomology). Furthermore, these relationships appeared to be bidirectional rather than unidirectional. Thus, neither language nor performance in the congruent condition appeared to be uniquely predictive of the other construct 1 year later. The interested reader may find these analyses and a brief summary in Supplemental Material S1.

Analysis

There are varied recommendations in the literature about which predictors are important in SLI and TD group comparisons, such as NVIQ or simple RT, particularly with regard to long-range relationships (i.e., compared to concurrent relationships; Kover & Atwood, 2013; Willoughby et al., 2018). Given our interest in analyzing long-range relationships, we first narrowed down the number of potential predictors of our outcome variables. Then, we incorporated information from the literature into our statistical models in order to increase the confidence by which we were able to draw meaningful conclusions. Therefore, our first analytical step was to use a model selection method that accounts for the uncertainty in choosing any specific model and, instead, averages over a large number of plausible models. Specifically, we used Bayesian model averaging (BMA; Raftery et al., 1997) to sort through all possible combinations of predictors that may have generated the data. We adopted a 90% posterior inclusion probability criterion ¹ in order to focus analyses on the most important Year 1 predictors of Year 2 performance. We used the “BMA” package in R (Raftery et al., 2018; R Core Team, 2018) and included the following covariates: Year 2 age, SES, Conners t scores, NVIQ, and baseline visual RT.

On the basis of variables that met our inclusion criterion in BMA analyses, we then conducted Bayesian linear regression using the “rjags” package in R (Plummer et al., 2018). Models included interaction terms for Group × Language or Inhibition predictor (e.g., Group × Morphology, Group × RT) and prior distributions derived from the literature (i.e., means, precision estimates; Kaplan, 2014). We report regression results using information from prior literature for all comparisons (see Supplemental Material S1 for comparison of informative vs. weakly informative priors). Our criterion for an effect to be considered important was that zero could not be contained within the 95% posterior probability interval ² (PPI). When a group interaction term did not meet this criterion, we did not analyze within-group relationships. We use this criterion in lieu of p values. Analyses conducted in the Bayesian framework do not yield frequentist-based p values as the conceptual underpinnings of these approaches differ. Rather than drawing a dichotomous conclusion based on evidence against the null hypothesis as in frequentist statistics, the Bayesian approach reveals to what degree the data support the null or alternative hypothesis (Kaplan, 2014).

To examine Year 2 nonreturners, we used typical group matching analyses and found no significant differences in child characteristics between Year 2 nonreturners and Year 2 returners (age: p = .24; SES: p = .08; NVIQ: p = .91; Conners t score: p = .26; Core Language: p = .43) or outcome measures (Morphology: p = .24; Vocabulary: p = .55; Flanker accuracy: p = .28; Flanker RT: p = .18) at Year 1. We also removed two participants from the original SLI group sample per regression diagnostics—leverage, regression model fit, and model influence (Judd et al., 2009).

Results

See Table 2 for descriptive performance, Tables 3 and 4 for summarized output, and Supplemental Material S1 for additional information on priors, model convergence, and complete output for all analyses.

Table 2.

Inhibition (Flanker incongruent condition) descriptive performance.

Inhibition measure	TD Year 1 (n = 48)	TD Year 2 (n = 41)	SLI Year 1 (n = 35)	SLI Year 2 (n = 30)
Reaction time (milliseconds)
M	616.316	575.445	691.514	616.526
SD	106.774	104.387	211.637	172.898
Accuracy (total correct / total trials)
M	0.950	0.954	0.865	0.903
SD	0.079	0.082	0.168	0.162

Note. TD = typically developing; SLI = specific language impairment.

Table 3.

Bayesian model averaging (BMA) output broken down by model (i.e., Year 2 outcome and Year 1 language or inhibition predictor) and group.

BMA model	Morph	Vocab	Acc	RT	Age	SES	NVIQ	Visual RT	Conners t score
Year 2 inhibition acc
TD morphology (N models = 18)
p≠	8.9				16.0	25.0	66.7	7.2	15.0
EV	−0.000				0.002	−0.002	−0.002	0.000	−0.000
SD	0.002				0.008	0.004	0.002	0.000	0.000
SLI morphology (N models = 20)
p≠	11.8				12.2	44.8	56.6	11.7	11.3
EV	−0.001				0002	0.006	0.003	−0.000	0.000
SD	0.007				0.012	0.009	0.004	0.000	0.001
TD vocabulary (N models = 18)
p≠		8.9			16.0	25.1	66.8	7.2	15.0
EV		−0.000			0.002	−0.002	−0.002	0.000	−0.000
SD		0.000			0.008	0.004	0.002	0.000	0.000
SLI vocabulary (N models = 22)
p≠		15.3			13.9	43.7	55.4	11.2	10.8
EV		0.000			0.002	0.006	0.003	−0.000	−0.000
SD		0.002			0.013	0.009	0.004	0.000	0.001
Year 2 inhibition RT
TD morphology (N models = 16)
p≠	10.0				47.7	11.4	12.9	77.4	9.6
EV	0.218				−13.943	−0.367	−0.137	0.589	0.006
SD	2.444				18.883	2.258	0.619	0.427	0.310
SLI morphology (N models = 19)
p≠	93.4				100.0	17.0	41.3	43.8	21.5
EV	−38.709				−103.989	0.330	−1.521	0.301	0.291
SD	18.128				27.965	3.093	2.444	0.456	0.957
TD vocabulary (N models = 17)
p≠		11.2			47.2	12.5	12.7	77.5	9.5
EV		0.033			−13.805	−0.441	−0.135	0.590	0.006
SD		0.364			18.847	2.454	0.615	0.428	0.308
SLI vocabulary (N models = 8)
p≠		100.0			100.0	17.9	14.0	18.2	90.6
EV		−9.463			−109.518	0.601	−0.139	0.054	3.784
SD		2.706			23.105	2.943	1.026	0.245	1.972
Year 2 morphology
TD accuracy (N models = 34)
p≠			72.5		14.8	27.0	63.8	48.1	12.6
EV			9.128		−0.054	0.064	0.057	0.008	−0.002
SD			7.399		0.258	0.144	0.055	0.011	0.014
SLI accuracy (N models = 26)
p≠			12.9		15.0	30.7	32.8	19.8	15.7
EV			0.079		0.034	0.042	0.018	−0.001	−0.003
SD			0.989		0.172	0.087	0.035	0.004	0.013
TD RT (N models = 34)
p≠				12.1	12.8	28.2	49.9	35.4	21.7
EV				−0.000	−0.020	0.071	0.042	0.005	−0.007
SD				0.002	0.216	0.154	0.053	0.010	0.020
SLI RT (N models = 20)
p≠				95.4	29.1	21.8	19.4	12.9	20.8
EV				−0.005	−0.153	0.019	0.006	−0.000	−0.004
SD				0.002	0.326	0.058	0.021	0.002	0.014
Year 2 vocabulary
TD accuracy (N models = 26)
p≠			12.8		48.8	98.7	33.0	27.1	57.3
EV			0.088		−2.183	2.875	0.099	0.016	−0.158
SD			10.780		2.841	1.011	0.191	0.037	0.175
SLI accuracy (N models = 23)
p≠			16.3		88.6	23.3	70.5	14.4	24.2
EV			1.591		−4.207	0.131	0.303	−0.003	0.035
SD			6.464		2.365	0.372	0.259	0.016	0.091
TD RT (N models = 30)
p≠				61.0	31.4	99.3	36.7	22.8	58.6
EV				0.032	−1.146	2.850	0.112	0.011	−0.157
SD				0.032	2.312	0.942	0.194	0.0311	0.169
SLI RT (N models = 20)
p≠				85.1	100.0	23.7	55.6	15.6	22.5
EV				−0.025	−7.345	0.110	0.191	−0.001	0.024
SD				0.016	2.407	0.326	0.225	0.014	0.075

Note. All models achieved 100% cumulative posterior probability; see above for the number of models selected by the model search algorithm. Results for the intercept are not shown; see Supplemental Material S1 for additional information. Morph/Morphology = morphological comprehension measured by Test of Language Development: Intermediate–Fourth Edition Morphological Comprehension subscale (Hammill & Newcomer, 2008); Vocab/Vocabulary = receptive vocabulary measured by the Peabody Picture Vocabulary Test–Fourth Edition (Dunn & Dunn, 2007); Acc = inhibition accuracy; RT = inhibition reaction time; SES = socioeconomic status measured by maternal education; NVIQ = nonverbal IQ measured by Wechsler Intelligence Scale for Children–Fourth Edition (Wechsler, 2003); Visual RT = baseline visual reaction time; Conners t scores = ADHD Rating Scale (Conners et al., 2008); TD = typically developing; p≠ = posterior inclusion probability; SLI = specific language impairment.

Table 4.

Bayesian regression output broken down by model (Year 2 outcome and Year 1 predictors), informative prior distributions (priors and sources are listed) versus weakly informative prior distributions, and within-group follow-up regression.

Regression model	B	SD	HPD interval	Penalized deviance	Bayes p value	Prior		Source
						TD	SLI
Year 2 inhibition RT
Year 1 morphology
Informative				8,765	.504	300ms	400ms	Kail, 1994; McElree, 2001; Schul et al., 2004; Van Dyke & Johns, 2012
Intercept	459.480	30.753	399.149, 528.983
Group	126.112	28.844	69.834, 181.687
Morphology	6.886	3.286	0.571, 13.398
Group × Morphology	−8.280	3.667	−15.454,−1.226
Age	5.462	3.409	−1.250, 11.953
Weakly informative				8,765	.501	0.001	0.001
Intercept	129.127	31.131	66.528, 187.482
Group	32.944	28.431	−23.607, 86.422
Morphology	14.986	3.578	7.863, 21.779
Group × Morphology	3.578	3.712	−3.265, 11.146
Age	29.920	3.742	22.606, 37.093
Within-group SLI Informative				3,870	.500		400ms	Kail, 1994; McElree, 2001; Schul et al., 2004; Van Dyke & Johns, 2012
Intercept	456.254	30.754	399.888, 521.376
Morphology	2.499	6.144	−9.081, 14.904
Age	11.672	4.763	2.270, 20.891
Within-group TD Informative				4,872	.504	300ms		Kail, 1994; McElree, 2001; Schul et al., 2004; Van Dyke & Johns, 2012
Intercept	353.063	30.454	295.712, 414.010
Morphology	13.608	3.464	7.141, 20.789
Age	6.773	4.277	−2.097, 14.801
Year 1 vocabulary
Informative				8,374	.499	300ms	400ms	Kail, 1994; McElree, 2001; Schul et al., 2004; Van Dyke & Johns, 2012
Intercept	422.653	30.595	362.083, 481.899
Group	119.542	30.775	59.664, 179.670
Vocabulary	1.699	0.434	0.840, 2.542
Group × Vocabulary	−0.677	0.352	−1.367, 0.022
Conners t score	1.701	0.437	0.832, 2.546
Age	−12.558	4.547	−21.751, −3.989
Weakly informative				8,785	.499	0.001	0.001
Intercept	89.805	30.708	29.898, 149.884
Group	17.712	30.764	−40.803, 79.515
Vocabulary	2.928	0.459	2.003, 3.800
Group × Vocabulary	0.408	0.358	−0.285, 1.112
Conners t score	2.082	0.455	1.190, 2.973
Age	3.983	4.749	−5.285, 13.241
Within-group SLI informative				3,857	.500		400ms	Kail, 1994; McElree, 2001; Schul et al., 2004; Van Dyke & Johns, 2012
Intercept	441.850	31.436	379.878, 502.911
PPVT	0.162	0.992	−1.807, 2.085
Conners t score	3.598	0.894	1.845, 5.353
Age	−11.294	7.183	−25.296, 2.898
Within-group TD informative				4,864	.500	300 ms		Kail, 1994; McElree, 2001; Schul et al., 2004; Van Dyke & Johns, 2012
Intercept	333.504	30.680	272.920, 393.641
PPVT	2.198	0.423	1.365, 3.027
Conners t score	0.862	0.479	−0.069, 1.817
Age	−6.822	5.502	−18.077, 3.662
Year 2 morphology
Year 1 inhibition RT
Informative				3,101	.501	11	7	Kaushanskaya et al., 2017; Norbury et al., 2016; Tomblin et al., 1997
Intercept	10.321	0.330	9.668, 10.968
Group	−2.034	0.658	−3.326, −0.760
Inhibition RT	−0.002	0.001	−0.003, −0.000
Group × Inhibition RT	−0.003	0.008	−0.005, −0.001
Within-group SLI informative				1,166	.500		7	Kaushanskaya et al., 2017; Norbury et al., 2016; Tomblin et al., 1997
Intercept	9.157	1.387	6.427, 11.917
Inhibition RT	−0.003	0.001	−0.004, −0.002
Age	0.011	0.111	−0.207, 0.232
Within-group TD informative				1,875	.500	11		Kaushanskaya et al., 2017; Norbury et al., 2016; Tomblin et al., 1997
Intercept	14.009	1.823	10.432, 17.649
Inhibition RT	−0.001	0.001	−0.002, 0.001
Age	−0.233	0.153	−0.538, 0.067

Note. All models included predictors that met our ≥ 90% posterior inclusion probability criterion in Bayesian model averaging analyses, and all priors had low precision (0.001) due to the novelty of analyses. HPD = highest posterior density; TD = typically developing; SLI = specific language impairment; RT = inhibition reaction time; Morphology = morphological comprehension measured by Test of Language Development: Intermediate–Fourth Edition Morphological Comprehension subscale (Hammill & Newcomer, 2008); Vocabulary = receptive vocabulary measured by the Peabody Picture Vocabulary Test–Fourth Edition (Dunn & Dunn, 2007); Conners t scores = ADHD Rating Scale (Conners et al., 2008).

Language as a Predictor of Inhibition

Flanker Incongruent Accuracy

BMA output indicated that no predictors met our ≥ 90% criterion in either the SLI or TD group model.

Flanker Incongruent RT

BMA output indicated that the following predictors met our ≥ 90% criterion in either the SLI or TD group model: age, Morphology, Conners t scores, and Vocabulary.

For the Morphology model, follow-up regression analysis indicated no important effects of age or Conners t scores. There were important effects of group (b = 126.112, SD = 28.844, 95% PPI [69.834, 181.687]), Morphology (b = 6.886, SD = 3.286, 95% PPI [0.571, 13.398]), and the interaction between group and Morphology (b = −8.280, SD = 3.667, 95% PPI [−15.454, −1.226]). There was an important effect of morphology for the TD group (b = 13.609, SD = 3.464, 95% PPI [7.141, 20.789]), but not for the SLI group (b = 2.499, SD = 6.144, 95% PPI [−9.081, 14.904]; see Figure 1), indicating that higher Morphology scores predicted higher later RT for the TD group.

Figure 1.

This figure presents the raw data and simple regression lines for Year 1 morphological comprehension as a predictor of Year 2 inhibition (Flanker incongruent condition) reaction time (RT). Statistical analyses indicated that zero was not contained in the typically developing (TD) group's credible interval and zero was contained in the specific language impairment (SLI) group's credible interval, indicating a relationship between Year 1 morphological comprehension and Year 2 inhibition RT for only the TD group. Data points are jittered.

For the Vocabulary model, follow-up regression analysis indicated that there were important effects of age (b = −12.558, SD = 4.547, 95% PPI [−21.751, −3.989]), Conners t scores (b = 1.701, SD = 0.437, 95% PPI [0.832, 2.546]), group (b = 119.542, SD = 30.775, 95% PPI [59.664, 179.670]), and Vocabulary (b = 1.699, SD = 0.434, 95% PPI [0.840, 2.542]), but not the interaction between group and Vocabulary (b = −0.677, SD = 0.352, 95% PPI [−1.367, 0.022]). Findings indicate higher Vocabulary scores predicted higher later RT for both groups.

Inhibition as a Predictor of Language

Morphology

BMA output indicated that the following predictor met our ≥ 90% criterion in either the SLI or TD group model: RT. For the RT model, follow-up regression analysis indicated important effects of group (b = −2.034, SD = 0.658, 95% PPI [−3.326, −0.760]), RT (b = −0.002, SD = 0.001, 95% PPI [−0.003, −0.000]), and the interaction between group and RT (b = −0.003, SD = 0.001, 95% PPI [−0.004, −0.001]). There was an important effect of RT for the SLI group (b = −0.003, SD = 0.001, 95% PPI [−0.004, −0.002]), but not the TD group (b = −0.001, SD = 0.001, 95% PPI [−0.002, 0.001]; see Figure 2), indicating that higher RT predicted lower later Morphology scores for the SLI group.

Figure 2.

This figure presents the raw data and simple regression lines for Year 1 inhibition (Flanker incongruent condition) reaction time (RT) as a predictor of Year 2 morphological comprehension. Statistical analyses indicated that zero was not contained in the specific language impairment (SLI) group's credible interval and zero was contained in the typically developing (TD) group's credible interval, indicating a relationship between Year 1 inhibition RT and Year 2 morphological comprehension for only the SLI group. Data points are jittered.

Vocabulary

BMA output indicated that the following predictors met our ≥ 90% criterion in either the SLI or TD group model: SES and age.

Discussion

The current study examined the predictive relationship between two indices of language—morphological comprehension and receptive vocabulary—and inhibition on a nonverbal task (Flanker) at two time points in an SLI versus TD group. Results indicated that neither language index at Year 1 predicted inhibition accuracy (i.e., incongruent condition) at Year 2, but both Year 1 language indices predicted Year 2 inhibition RT (i.e., incongruent condition). This predictive relationship differed between groups for morphological comprehension, but not receptive vocabulary. For TD children, Year 1 morphological comprehension predicted Year 2 inhibition RT. This relationship was not present in the SLI group. Results also indicated Year 1 accuracy did not predict either Year 2 language index and Year 1 inhibition RT predicted morphological comprehension, but not receptive vocabulary. This predictive relationship differed across groups. For children with SLI, Year 1 inhibition RT predicted Year 2 morphological comprehension. This relationship was not present in the TD group. This finding suggests that children with SLI who are faster to inhibit irrelevant information have better morphological comprehension 1 year later relative to children with SLI who are slower to inhibit irrelevant information.

Predictive Power of Language Versus Inhibition

The current findings corroborate Dispaldro et al.'s (2013) finding of a group difference in the concurrent relationship between receptive grammar and inhibition yet diverge from Dispaldro et al. because this relationship was only present in their SLI group. Our findings also diverge from studies that find weak or no significant correlations between language and inhibition (Bishop & Norbury, 2005; Ladányi & Lukács, 2016). Discrepancies may relate to differences in outcome measures as Dispaldro et al. only examined accuracy, rather than both accuracy and RT as in the current study. They may also relate to the particular index of language and participant group examined. Kaushanskaya et al. (2017) found unique associations between a composite syntax measure, but not a composite vocabulary measure, and inhibition on a nonverbal task in TD children, whereas Bishop and Norbury (2005) found only weak correlations between a composite vocabulary and syntax measure, and inhibition on verbal and nonverbal tasks when combining SLI and TD groups. Taken together, the current study supports prior work that demonstrates a relationship between language and inhibition and extends prior work by demonstrating group differences in the relationship between morphological comprehension and inhibition RT.

The current study examined directionality of influence between language and inhibition. Year 1 receptive vocabulary and morphological comprehension predicted Year 2 inhibition RT across groups, but neither Year 1 inhibition accuracy nor RT predicted Year 2 receptive vocabulary. Year 1 inhibition RT predicted Year 2 morphological comprehension in the SLI group, but Year 1 morphological comprehension predicted Year 2 inhibition RT in the TD group. These results converge on findings from prior work showing a stronger relationship between morphological comprehension and inhibition than receptive vocabulary and inhibition (Dispaldro et al., 2013; Kaushanskaya et al., 2017) and other work positing a close relationship between inhibition and morphosyntax (Ibbotson & Kearvell-White, 2015; Marton et al., 2018; Miller et al., 2001). Finally, results are broadly consistent with prior findings from our lab on a complex EF task with minimal verbal demands. Larson et al. (2019) showed that children with SLI and relatively better language ability had faster response times than children with SLI and relatively poorer language ability. We also showed that this relationship was reversed in TD peers.

Theoretical and Clinical Implications

The relationships we found between language and inhibition in the TD group align with the HCSM as language played an important role in this EF skill—potentially signaling their use of language to reflect on behavior (Marcovitch & Zelazo, 2009). SLI group patterns, however, do not appear to align with the HCSM as this EF skill played an important role in later language. On the surface, the link between inhibition RT (i.e., incongruent condition) and later morphological comprehension in SLI aligns with the generalized slowing hypothesis (Kail, 1994; Miller et al., 2001), but this link exists even when statistically accounting for baseline visual RT. Alternatively, the current findings provide evidence for the inefficient inhibition hypothesis (Bjorklund & Harnishfeger, 1990, 1995; Marton et al., 2007)—inhibition plays a key role in language, and SLI deficits in language may relate to inhibition weaknesses. Thus, the tension between processing speed and inhibitory control appears to be important to morphosyntactic development in SLI.

Vocabulary appears to influence later inhibition across the SLI and TD groups. Morphology also appears to influence later inhibition in the TD group; however, inhibition appears to influence later morphology in the SLI group. Thus, the relationship between language and inhibition differs between the SLI and TD groups for a morphology index, but not a vocabulary index. This evidence aligns with accounts of procedural and statistical learning deficits in children with SLI, which suggest underlying difficulty with the long-range contingencies necessary for morphological processing (Hsu & Bishop, 2011; Obeid et al., 2016; Ullman & Pierpont, 2005).

There are a few preliminary clinical implications of the current study. It is possible that targeting inhibition skills as a part of language intervention may improve subsequent morphological comprehension—perhaps by freeing up WM resources to process a greater proportion of relevant to irrelevant information. Improvements in inhibition may be associated with diminished morphosyntactic deficits over time based on the current findings. However, additional measures of language and inhibition are indicated prior to clinical application, such as naturalistic language and inhibition assessment.

Limitations and Future Directions

Although groups were matched on age, group differences were present on baseline visual RT, SES, and NVIQ. These differences are consistent with prior work (Kover & Atwood, 2013; Willoughby et al., 2018), and our analytical approach allowed us to examine how likely these factors were to exist in the true model of performance (Hoeting et al., 1999; Raftery et al., 1997). Returner and nonreturner groups did not differ on participant characteristics, yet it is possible that we did not account for all potential characteristics that relate to performance.

Additional time points and measures of language and inhibition across differing modalities (e.g., auditory vs. visual inhibition) are also necessary to support and extend current findings. For instance, our morphology assessment involved both inflectional and derivational morphemes, which have different developmental progressions; thus, it may be useful to examine relationships between inhibition and specific grammatical morphemes or syntactic constructions. Our vocabulary assessment was more heavily focused on nouns than other types of words, and it may be useful for future work to examine relationships between inhibition and a greater breadth of vocabulary (e.g., question words, prepositions). Designing experimental tasks of language may also enhance the ability of future work to, for instance, relate long-range syntactic contingencies with inhibition. It will also be important to investigate causality at earlier stages of development. Furthermore, we examined trial-level performance on the Flanker congruent and incongruent conditions separately rather than an aggregated difference score. It may be beneficial for future work to examine other measures drawing on inhibitory functions (e.g., switching, planning) to extend current findings and rule out attentional processes as a predictor of performance beyond ADHD symptomology and baseline RT.

Conclusions

We examined the Year 1 to Year 2 predictive relationships between two language indices—morphological comprehension and receptive vocabulary—and inhibition on a nonverbal task in an SLI versus TD group. We found a relationship between language and inhibition, but this relationship differed between groups for morphological comprehension. For the SLI group, inhibition RT played a larger predictive role in later morphological comprehension relative to inhibition accuracy and relative to language in later inhibition performance. These findings provide an initial step in characterizing causal relationships between language and nonlinguistic cognitive processes in school-age children that can inform future research. Our findings provide preliminary support that improvements in inhibition skills in children with SLI may attenuate their pervasive deficits in morphosyntax.

Funding Statement

This research was supported by National Institutes of Health Grants R01 DC011750 (Ellis Weismer and Kaushanskaya, multiple principal investigators), T32 DC005359 (Ellis Weismer, principal investigator), and U54 HD03352 core grant to the Waisman Center.