1. Introduction
The current investigation concerns a short and engaging adaptive working-memory training intervention that aimed to improve both language comprehension and working memory in children with developmental language disorder. As well as looking at ‘far transfer’ to two measures of language comprehension (sentence comprehension and receptive grammar), the study included six measures of ‘near transfer’ to assess whether the intervention led to gains in working-memory tasks that were not directly trained.
The working-memory system is widely accepted as describing the way information is processed and retained [
1] and can be seen as a mental workspace which encompasses several skills used for ‘online’ activities during daily life [
1,
2]. There are several approaches to understanding working memory [
3,
4,
5,
6]; one of the most influential is Baddeley and Hitch’s [
7] ‘Multi Component Model of Working Memory’, and its subsequent revision by Baddeley [
8]. This model consists of several interacting components, including a limited attentional capacity control system, termed the central executive, which is assisted by two further ‘passive’ components, the phonological loop, for holding speech-based information, and the visuospatial sketchpad, for holding visual and spatial information [
9,
10]. The central executive relies heavily, but not exclusively, on the frontal lobes [
11], and is actively responsible for attentional control, directing and allocating resources in activities when there is a need for the retention of information together with the processing of other information [
9,
10]. We use the term ‘executive working memory’ (EWM) to refer to working-memory assessments that involve input from the central executive (e.g., concurrent storage and processing) and to interventions that target this component. There also is a fourth component, the episodic buffer, that provides links to long-term memory and some additional multi-modal storage, but this component will not be considered further.
Researchers have emphasised that understanding working-memory development is crucial for helping children maximize their intellectual progress [
12]. There are significant associations between the components of the working-memory system and other cognitive processes, especially those related to language [
13], literacy [
14] and intelligence [
15,
16] (but see [
17]). It is notable that many groups of children with developmental conditions have significantly lower abilities when assessed on the components of working memory compared to children with typical development (TD) who have a similar chronological or sometimes even mental age [
1]. Examples of this are: language disabilities [
18], dyslexia [
19], intellectual disabilities [
20], and autism [
21].
These associations between working memory and other cognitive processes, as well as the lower working-memory performance of children with developmental conditions, underpin the interest in using interventions to improve working memory and related neuropsychological abilities. An investigation of a computerised working-memory intervention targeting the central executive found that, following training, performance on components of the working-memory system improved, and importantly, non-verbal intelligence also improved [
22]. These effects were reported for children with attention deficit hyperactivity disorder and for a small sample of adults with TD [
22]. This initial study was followed by further investigations of this claim, which have been summarised in subsequent systematic reviews and meta-analyses. In general, these reviews have concluded that training can result in near transfer, involving an increase in other working-memory abilities that were not part of the intervention procedure, although the evidence for the persistence of these effects is variable. The reviews have also concluded that interventions generally do not result in far transfer to other cognitive abilities such as those associated with mathematics, intelligence, literacy, or language.
However, systematic reviews to date have usually been concerned with computer-based interventions that attempt to improve at least one component of the working-memory system [
23,
24,
25,
26] or non-computerised interventions involving mainly children with typical development [
27]. Consequently, although these reviews suggest that working-memory training does not result in far transfer, their focus has not been on non-computerised interventions targeting children with developmental conditions.
It remains important to understand how working-memory interventions might benefit children with developmental conditions. Rowe et al. [
27] emphasised the potential importance of exploring working-memory training interventions that target executive skills with children who have language or other disabilities. This is particularly important as a meta-analysis by Peijnenborgh et al. [
28] of working-memory interventions with children who have developmental conditions showed that previous research had focussed mainly on children with attention deficit hyperactivity disorder (10/13 studies), or the type of developmental condition was not specified. In addition, several investigations since 2018 challenge the pessimistic conclusions reached in the earlier reviews of this topic. For example, a single case experimental design investigation involving seven children with working-memory difficulties [
29] used the Cogmed computer-based intervention [
30]. This study found near-transfer effects for all children, and far transfer in language, reading, or mathematics for three children with ‘convincing but modest training effects across multiple measures’ [
22] (p. 1).
Several recent investigations of working-memory interventions have concerned children with developmental language disorder (DLD) or a related diagnosis. DLD is a persistent neurodevelopmental condition, involving a primary difficulty with language that is unexplained by any other syndrome or circumstances [
31,
32]. DLD is estimated to have a prevalence of 7% [
33].
Newly emerging evidence suggests a more promising role for working-memory interventions in children with DLD. For example, in a study using a pre-test/intervention/post-test design, Holmes et al. [
34] used the Cogmed working-memory intervention with a sample of children who had low language ability. They identified significant training gains for several working-memory components and a far-transfer effect to performance IQ, although this effect was non-significant with Bonferroni corrections. Acosta, Hernandez and Ramirez [
35] also used a pre-test/intervention/post-test design, developing a working-memory training which targeted short-term verbal memory and EWM in both verbal and visuospatial domains, delivered using a combination of individual and group sessions. They found significant training gains in all components of the working-memory system for children with language difficulties, as well as far transfer to a lexical–semantic task; they also noted greater gains in their group with language difficulties than in a comparison group with TD. However, neither of these studies included an untreated or active control group, which would provide stronger evidence in favour of the intervention [
24].
Acosta-Rodríguez, Ramírez-Santana, and Hernández-Expósito [
36] used a comprehensive multi-technique non-computer classroom-based intervention that included executive working-memory training as one element. They included children with DLD and TD, allocating half of the children in each group to receive the intervention and half to receive no treatment. Acosta-Rodríguez and colleagues [
36] reported immediate post-intervention improvement on several assessments of language comprehension, verbal EWM and semantic fluency in trained children with DLD compared to the untreated DLD control group. The children with DLD receiving the intervention made greater gains than the equivalent typical group. In a single case experimental design, Maleki Shahmahmood et al. [
37] used a mixed non-computer and computer-based intervention with 10 Iranian children who were identified with a primary language impairment. The intervention focused on phonological short-term memory and EWM, as well as suggestions about strategies. Following training over five weeks, increases in working memory and morpho-syntactic abilities were reported. A subsequent five-week language intervention also increased morpho-syntactic abilities, but gains were not detected for working memory, suggesting an effect of working-memory training on language, but not vice versa [
37].
In two other investigations, far-transfer effects of working-memory training have been reported on syntax with French children who have DLD [
38,
39]. Stanford et al. [
38] used iPad or computer-based training (Magic Memory) in groups with DLD and typical development (TD). They also included active control groups who followed an alternative general scholastic training without working-memory content, and randomly allocated their participants to each condition. The training process involved practising five verbal working-memory tasks assessing two components of the working-memory system (verbal short-term memory and EWM). These working-memory training tasks were adaptive, such that as the child’s performance improved or declined, the task difficulty was adjusted to match the child’s current performance. Improvements in both aspects of working memory were found immediately post-intervention for DLD and TD groups. Furthermore, for the DLD group there were improvements in syntax involving third-person accusative clitics, which are often seen as a marker for DLD in French-speaking children [
38]. Delage et al. [
39] used the same Magic Memory intervention, reporting improved working memory in DLD and TD groups compared to active control groups immediately post-intervention, with additional far-transfer gains on sentence repetition and complex syntax in the DLD intervention group. Therefore, the findings reported since 2018 are not only promising support for working-memory interventions for children with DLD, but also suggest that children with DLD may gain more from working-memory interventions compared to individuals with TD [
24,
34,
35,
36,
38,
39,
40,
41].
The present investigation of the effectiveness of an executive working-memory training intervention for children with DLD builds on these findings. We focussed on children with DLD because of the suggestion that children with developmental conditions may be more likely to benefit from working-memory interventions, as there is more room for improvement. Children with DLD are known to have difficulties with working memory, which encompass both short-term and executive working memory [
35,
42,
43,
44,
45]. A face-to-face rather than a computer-based intervention was chosen because, when the study was planned, meta-analyses suggested computer-based forms of intervention were not effective [
24,
26]. The intervention targeted EWM skills, based on previous evidence that this is the common feature of effective working-memory interventions [
27], and evaluated far-transfer effects to receptive grammar and language comprehension skills. It was predicted that the intervention would improve receptive grammar abilities, given evidence reported above that working-memory interventions can improve morpho-syntax in children with DLD, possibly by increasing information-processing abilities. It was also predicted that the intervention would improve language comprehension, based on previous findings and work by Diamond [
46] and others [
47] who emphasise the strong links between language comprehension and EWM, both of which require the concurrent processing and retention of continuous input that unfolds over time. In this context, Melby-Lervåg et al.’s [
24] report of a small significant effect of working-memory training on reading comprehension (and arithmetic) is interesting, although when they excluded studies that had a significant decline pre- to post-intervention in relevant control groups, the effect was not regarded as noteworthy.
The current intervention followed a programme used by Henry et al. [
48], who reported promising longer-term findings for reading comprehension in children with TD. This programme comprised a relatively short, face-to-face, enjoyable, adaptive (an important feature of effective interventions [
49]) intervention targeting EWM skills. The intervention employed both verbal and non-verbal EWM tasks as recommended by Danielsson et al. [
50], who only found significant training effects to untrained working-memory measures in samples with a developmental condition (intellectual disabilities) when a mixed working-memory approach was adopted that included both verbal and visuo-spatial components. In the current study, participants were assessed before intervention, and re-assessed immediately post-intervention and nine months later, evaluating maintenance over a longer period than any previous study of children with DLD.
The overall research question was: For children with DLD, are there immediate and longer-term effects of EWM training? We assessed immediate and longer-term effects for directly trained EWM tasks, working-memory tasks that were not trained (near transfer) and language comprehension (far transfer).
3. Results
3.1. Pre-Test Scores
Mean raw scores for pre-intervention, post-intervention and 9-month follow-up on all outcome measures are given in
Table 2. All scores for working-memory tasks represent total trials correct as these scores were comparable across all tasks (standardised and non-standardised), and in some cases, standardised scores were unavailable due to low performance. For Sentence Comprehension, we report raw scores; and for Receptive Grammar, total blocks passed.
There were no significant differences between groups on any pre-test scores (see
Table 2 for means and SDs):
Listening Recall, t(45) = −0.23, p = 0.819 (d = −0.07, 95%CI −0.60, 0.51);
Odd One Out, t(45) = 0.28, p = 0.782 (d = 0.08, 95%CI −0.49, 0.65);
Sentence Comprehension, t(45) = 1.17, p = 0.25 (d = 0.34, 95%CI −0.24, 0.91);
Receptive Grammar, t(45) = 1.62, p = 0.113 (d = 0.47, 95%CI −0.11, 1.05);
Digit Recall, t(45) = 0.82, p = 0.415 (d = 0.24, 95%CI −0.34, 0.81);
Word List Recall, t(45) = 1.90, p = 0.064 (d = 0.56, 95%CI −0.03, 1.14);
Block Recall, t(45) = 0.37, p = 0.711 (d = 0.11, 95%CI −0.46, 0.68);
Pattern Span, t(45) = 0.74, p = 0.465 (d = 0.22, 95%CI −0.36, 0.79);
Counting Recall, t(45) = 0.80, p = 0.427 (d = 0.23, 95%CI −0.34, 0.80);
Backwards Digit Recall, t(45) = 0.87, p = 0.389 (d = 0.25, 95%CI −0.32, 0.83).
Scores for Sentence Comprehension and Receptive Grammar were numerically lower in the active control group, in line with the language screening scores. Therefore, all analyses of outcome measures controlled for pre-test performance so that any differences between groups could be attributed to the intervention and not initial levels of performance [
58].
3.2. Approach to Analyses
Hierarchical multiple regression was undertaken to examine whether training group (working-memory training vs. active control) was a significant predictor of post-intervention and 9-month follow-up outcome scores. In all regressions, pre-intervention scores for the relevant outcome measures and age in months were controlled at step one. Controlling for pre-intervention outcome scores was a conservative approach as it ensured that any initial variations in performance across participants and groups were taken into account [
58]. Age was also controlled as we were using raw scores uncorrected for age. The intervention group, as a dummy variable, was entered at step two of each regression to assess the effects of the intervention.
For all regression analyses, key statistical checks were carried out (Durbin–Watson, tolerance and VIF statistics, Cook’s and Mahalanobis distances, standardised DFBeta, leverage values, plots of standardised residuals, predicted standardised values, standardised residuals and partial plots). In some regression analyses, one child in the working-memory training group improved on an outcome measure more than the model predicted, resulting in a high standardised residual (outlier). When this occurred, regressions were re-run after removing the relevant case, but as there were never any substantive differences in findings, original analyses are reported.
3.3. Direct Effects (Listening Recall and Odd One Out)
The first analyses assessed whether training intervention effects were present on the two trained executive working-memory tasks. As four separate regression analyses were conducted (two directly trained executive working-memory tasks at two time points), Bonferroni corrections were made to the significance levels for the overall models (
p < 0.01). See
Table 3 for summary regression information and
Figure 2 for graphs illustrating pre-intervention, post-intervention and 9-month follow-up scores for each group.
For post-intervention Listening Recall, at step one of the regression, the pre-intervention scores and age accounted for 44.3% of the variance. Examination of the standardised beta-values indicated that only the pre-intervention score was a significant individual predictor; this was not unexpected as pre-intervention scores are an important predictor for scores after intervention. In step two, adding group produced a significant change in R2 (∆R2); here, group accounted for a further 33.8% of the variance. The final model was significant, F(3, 43) = 51.06, p < 0.001, predicting 78.1% of the total variance (Adj. R2 0.766). At 9-month follow-up, pre-intervention scores and age accounted for 44.9% of the variance at step one, with only pre-intervention score as a significant individual predictor. Adding group at step two again produced a significant change in R2 (∆R2), with group accounting for a further 35.8% of the variance. The final model was significant, F(3, 43) = 59.89, p < 0.001, predicting 80.7% of the total variance (Adj. R2 0.793).
The training group differences on Listening Recall were large, even after controlling for pre-intervention score and age. Inspection of unstandardised beta values (B) in
Table 3 shows that, holding age and pre-intervention score constant, being in the active control group rather than the working-memory training group meant that the Listening Recall trials correct scores were on average 6.61 points lower (effect size
f2 = 0.51) at post-intervention and 6.85 points lower (
f2 = 0.56) at 9-month follow-up. Note: for Cohen’s
f2 [
59], the effect sizes are
f2 = 0.02 small;
f2 = 0.15 medium;
f2 = 0.35 large.
For post-intervention Odd One Out, at step one, the pre-intervention scores and age accounted for 30% of the variance, with the pre-intervention score representing a significant individual predictor. Adding group at step two produced a significant change in R2 (∆R2), accounting for a further 48.3% of the variance; here, age became a significant predictor alongside pre-intervention score and group. The final model was significant, F(3, 43) = 51.58, p < 0.001, predicting 78.3% of the total variance (Adj. R2 0.767). At 9-month follow-up, pre-intervention scores and age accounted for 31% of the variance, with pre-intervention score a significant individual predictor. Adding group at step two produced a significant change in R2 (∆R2), accounting for a further 47.2% of the variance; here, age again became a significant predictor alongside pre-intervention score and group. The final model was significant, F(3, 43) = 51.47, p < 0.001, predicting 78.2% of the total variance (Adj. R2 0.767).
The training group differences for Odd One Out were very large, even after controlling for pre-intervention score and age. Inspection of unstandardised beta values in
Table 3 shows that, holding age and pre-intervention score constant, being in the active control group rather than the working-memory training group, meant that Odd One Out trials correct scores were on average 9.70 points lower (effect size
f2 = 0.93) at post-intervention and 9.04 points lower (
f2 = 0.90) at 9-month follow-up.
Summary
The working-memory intervention led to substantial and significant gains in performance for those in the working-memory training group on two directly trained tasks, Listening Recall and Odd One Out span, with very large effect sizes. These findings were present at both time points and were found even after controlling for pre-intervention performance and age.
3.4. Far-Transfer Effects
We next assessed whether there were far-transfer effects to Sentence Comprehension and Receptive Grammar skills, either immediately post-intervention or at 9-month follow-up. As four separate regression analyses were conducted (two far-transfer variables, two time points), Bonferroni corrections were made to the significance levels for the overall models (
p < 0.01). See
Table 4 for summary regression statistics and
Figure 2 for graphs illustrating pre-intervention, post-intervention and 9-month follow-up scores for each group.
For post-intervention Sentence Comprehension, at step one, pre-intervention scores and age accounted for a substantial 68.5% of the variance, with only pre-intervention score a significant individual predictor. Of particular interest was the effect of adding group in step two, which produced a significant change in R2 (∆R2), accounting for a further 10.4% of the variance. The final model was significant, F(3, 43) = 53.61, p < 0.001, predicting 78.9% of the total variance (Adj. R2 0.774). At 9-month follow-up, pre-intervention scores and age again accounted for a substantial 66.2% of the variance at step one, with only pre-intervention score a significant individual predictor. Again, in step two, adding group produced a significant change in R2 (∆R2), accounting for a further 9.7% of the variance. The final model was significant, F(3, 43) = 45.08, p < 0.001, predicting 75.9% of the total variance (Adj. R2 0.742).
Thus, even with our conservative approach of controlling for the substantial variance accounted for by pre-intervention scores and age, training group was a significant predictor of Sentence Comprehension both immediately and nine months following the intervention. Inspection of unstandardised beta values in
Table 4 shows that, when holding age and pre-intervention score constant, being in the active control group rather than the working-memory training group meant that Sentence Comprehension scores were on average 3.98 points lower (effect size
f2 = 0.12) at post-intervention and 3.60 points lower (
f2 = 0.11) at 9-month follow-up. The gains after training approached a medium effect size. If we consider the average pre-intervention Sentence Comprehension score in our working-memory training group (raw score = 21), together with their average age (8 years 0 months), a raw score of this value gives a scaled score of 6 according to the manual [
55]. An increase of 4 (raw score) points over the control group at post-intervention following training would result in a raw score of 25, with an increase in the scaled score from 6 to 9 (a jump from 9th to 37th percentile).
For post-intervention Receptive Grammar, pre-intervention scores and age accounted for 54% of the variance in post-intervention scores at step one, with pre-intervention score a significant individual predictor. In step two, adding group did not produce a significant change in R2 (∆R2), accounting for only a further 2.5% of the variance. The final model was significant, F(3, 43) =18.59, p < 0.001, predicting 56.5% of the total variance (Adj. R2 0.534). At 9-month follow-up, pre-intervention scores and age accounted for 58.1% of the variance at step one, with pre-intervention score a significant individual predictor. In step two, adding group, failed to produce a significant overall change in R2 (∆R2), accounting for a further 3.5% of the variance. The final model was significant, F(3, 43) = 22.93, p < 0.001, predicting 61.5% of the total variance (Adj. R2 0.589).
The training group was not a significant predictor of Receptive Grammar at post-intervention or 9-month follow-up, although children in the working-memory training group did obtain numerically higher scores. Inspection of unstandardised beta values in
Table 4 shows that, holding age and pre-intervention score constant, being in the active control group rather than the working-memory training group meant that blocks passed scores on Receptive Grammar were on average 1.40 points lower (effect size
f2 = 0.03) at post-intervention and 1.50 points lower (
f2 = 0.04) at 9-month follow-up (neither group difference was significant). In the latter case, however, there was a marginally significant effect of group (
p < 0.055) with a small effect size.
Summary
The working-memory intervention led to significant gains for those in the working-memory training group on Sentence Comprehension (three to four points in the raw score), which approached a medium effect size. These findings were evident both immediately post-intervention and at 9-month follow-up and were found even after controlling for pre-intervention performance and age. No significant training group differences were found for Receptive Grammar.
3.5. Near-Transfer Effects (Untrained Measures of Working Memory)
Also of interest was the possibility of near-transfer effects to related working-memory tasks that were not directly trained, both immediately after the intervention and at 9-month follow-up. We assessed four short-term working-memory measures (Digit Recall, Word List Recall, Block Recall, and Pattern Span) and two executive working-memory measures (Counting Recall and Backwards Digit Recall). As 12 separate regression analyses were conducted (six working-memory variables, two time points), Bonferroni corrections were made to the significance level for the overall models (
p < 0.004). See
Table 5 for regression summaries for the short-term memory measures and
Table 6 for regression summaries for the executive working-memory measures. See also
Figure 2 for graphs illustrating pre-intervention, post-intervention and 9-month follow-up scores for each group.
For the four short-term working-memory measures, all regression models at post-intervention and 9-month follow-up were significant (Fs(3, 43) ranged between 20.24 and 54.00, all ps < 0.001). Pre-intervention scores and age accounted for between 40.6% and 67.4% of the variance at step one in all eight models, and only pre-intervention scores were significant individual predictors in each case. At step two, adding group produced a significant change in R2 (∆R2) in all eight models, accounting for additional variance of between 10.8% and 24.0%. Being in the working-memory training group was associated with significantly higher scores on all four short-term memory assessments at both time points. The total variance accounted for in the models ranged from 58.5% to 79.0%.
Group differences associated with being in the working-memory training group were medium to medium-large across all near-transfer short-term memory measures, even after controlling for pre-intervention scores and age. Inspection of unstandardised beta values in
Table 5 shows that, holding age and pre-intervention score constant, being in the active control group rather than the working-memory training group meant that trials correct scores on short-term working-memory tasks were on average between 3.73 and 5.06 points lower at post-intervention (
f2 = 0.22, 0.24, 0.14, and 0.13 for Digit, Word List, Block and Pattern) and between 3.90 and 6.29 points lower at 9-month follow-up (
f2 = 0.31, 0.30, 0.20, and 0.12 for Digit, Word List, Block and Pattern).
For the two untrained executive working-memory measures, all regression models at post-intervention and 9-month follow-up were significant (Fs(3, 43) between 36.87 and 56.15, all ps < 0.001). Pre-intervention scores and age accounted for between 62.2% and 71.0% of the variance at step one, and only pre-intervention scores were significant individual predictors in each case. At step two, adding group produced a significant change in R2 (∆R2) in all models, accounting for additional variance of between 7.5% and 12.9%. For both EWM tasks at both time points, being in the trained group was associated with significantly higher scores. The total variance accounted for in these models ranged from 72.0% to 79.6%.
Group differences associated with being in the working-memory training group were medium to medium-small for the near-transfer executive working-memory measures, even after controlling for pre-intervention scores and age. Inspection of unstandardised beta values in
Table 5 shows that, holding age and pre-intervention score constant, being in the active control group rather than the working-memory training group meant that trials correct scores on Counting Recall and Backwards Digit Recall were on average 5.44 and 3.38 points lower respectively at post-intervention (
f2 = 0.15 for Counting Recall and
f2 = 0.08 for Backwards Digit Recall); and 5.34 and 3.17 points lower at 9-month follow-up (
f2 = 0.15 for Counting Recall and
f2 = 0.08 for Backwards Digit Recall).
Summary
The working-memory intervention led to significant gains for those in the working-memory training group on all six untrained working-memory tasks, encompassing both short-term and executive working memory, usually with medium effect sizes. This provided strong evidence for near transfer to similar working-memory tasks. These findings were evident both immediately post-intervention and at 9-month follow-up and were found even after controlling for pre-intervention performance and age.
4. Discussion
This study investigated whether an adaptive intervention targeting executive working memory would benefit children with developmental language disorder, as evidenced by improvements not only on the two trained tasks (direct effects) and six untrained working-memory tasks (near-transfer effects), but also on tests of Sentence Comprehension and Receptive Grammar (far-transfer effects). Outcomes were measured immediately post-intervention and nine months later. To specifically evaluate the contribution of an executive working-memory training intervention, the study randomly assigned participants to a ‘trained’ group who received the executive verbal and visuospatial tasks (Listening Recall and Odd One Out) and an ‘active control’ group who received the same tasks but without the executive element. Hierarchical multiple regression analyses controlling for pre-intervention performance and age found group to be a significant predictor of Sentence Comprehension at both time points. These findings indicated that children in the working-memory training group obtained significantly higher scores on Sentence Comprehension than children in the active control group immediately following training, and that these gains were maintained nine months later. We estimated how the intervention affected the children’s percentile scores on Sentence Comprehension. Calculations suggested that the intervention could result in a change from the 9th to 37th percentile, an appreciable gain, as after the intervention scores would reach the low-typical range. Performance on Receptive Grammar, however, did not show these gains, with only a marginal group effect at 9-month follow-up.
In terms of direct and near-transfer effects, group was a significant predictor of performance on both the trained executive working-memory tasks (Listening Recall and Odd One Out) and performance on all other near-transfer working-memory tasks (Digit Recall, Word List Recall, Block Recall, Pattern Span, Counting Recall and Backward Digit Recall) immediately post-intervention and nine months later. For all tasks, children in the working-memory training group obtained significantly higher scores than children in the active control group at both time points, even after controlling for pre-intervention performance and age.
The findings of positive near and far-transfer effects of the working-memory training in this study run counter to earlier systematic reviews that have concluded that working-memory interventions fail to produce far-transfer effects. However, these reviews were mostly based on studies of children with TD [
23,
24,
25,
26,
27]. In contrast, our findings add to more recent studies reporting far transfer to language performance, usually with children who have developmental conditions such as language difficulties [
29,
35,
36,
37,
38,
39]. It is notable that the present study achieved these positive outcomes following training on just two executive working-memory tasks (one verbal and one visuospatial) administered in 18 sessions of around 10 min three times a week (total 3 h). Other studies of children with DLD deployed more intervention tasks and included both short-term memory and executive working-memory components (e.g., five computer-based short-term and executive working-memory tasks in the Magic Memory programme [
38,
39] or multiple live memory tasks [
35]; and total intervention times ranging from 12–18 h across studies). The present study is also unique in conducting a longer-term follow-up and finding that gains were retained nine months after the short, targeted intervention.
How might we account for the positive near and far-transfer effects of this intervention? Face-to-face delivery was used on the grounds that social engagement may increase children’s motivation and focused attention to input, providing favourable conditions for take-up of the training. This may be particularly important for the executive-load element of EWM tasks that the trained group received, differentiating this from the judgement subtasks that the active control group received. Indeed, children in the present study enjoyed taking part, all willingly completing the 18 sessions, and children in the trained group were keen to improve their previous score in both tasks, in effect competing with themselves. However, far transfer to language performance has also been observed in recent studies using computer-based training that, like the present training, appears to turn relatively pure working-memory tasks into engaging activities [
38,
39]. This suggests that the focus and/or enjoyment of the activity and motivation rather than mode of delivery may be a key factor in positive outcomes. In addition, the adaptive nature of the executive task, ensuring children are largely successful, while encouraging them to perform just above their current level of success, may be an important factor in their engagement and motivation and is noted by many authors as a key ingredient for the success of working-memory interventions [
38,
39].
The effect of group on performance both post-intervention and nine months later indicates that the EWM element was the active ingredient in the intervention, since the active control group experienced the same input and judgement task as the working-memory training group but without the executive load that also required recall of the final word in each sentence (verbal intervention) or location of the ‘odd one out’ (visuospatial intervention). This outcome is in line with the prediction that an EWM training would benefit language comprehension on the grounds that both cognitive activities require simultaneous processing and storage of information [
46,
47]. In sentence comprehension, words need to be retained in short-term memory while also processing their meaning and the meaning of the sentence. Furthermore, most items in the Sentence Comprehension task used in this study required the child to select from a set of pictures the one that matched a spoken sentence (apart from the final, most difficult items that required inferencing from a sequence of sentences presented with pictorial support). To succeed, the child must not only recognise, understand, and retain all elements of the verbal input, but must at the same time scan the visual input to find the picture that depicts all elements of the sentence, avoiding the distraction of pictures that share some but not all of these elements. The implication is that the intensive experience of simultaneous processing and storage in the training tasks improved executive management, as evidenced by the post-training gains on these tasks, to the benefit of the executive-loaded Sentence Comprehension.
It is notable that the other measure of far transfer to language comprehension (Receptive Grammar from the Test for the Reception of Grammar, TROG-2 [
56]), did not show significant improvement in the working-memory training group, although there was a trend in this direction at 9-month follow-up. On our proposed interpretation of training effects, we might infer that the gains in executive management following training were not sufficient to meet the executive demands of the TROG-2 verbal/visual material, or that the TROG-2 includes grammatical structures that children have not acquired. These possibilities are plausible since the blocks of items testing grammatical constructs on the TROG-2 become increasingly complex both conceptually and syntactically. Another possibility is that the very stringent scoring on the TROG-2, with children only receiving a point for a grammatical construct if they respond correctly to all four items in the block targeting that construct, raised the bar for achieving statistically significant changes in scores.
The attribution of gains in Sentence Comprehension to improved EWM is in line with the account of training-induced changes in working memory put forward by Gathercole et al. in a systematic review of working-memory training effects [
60]. According to their account, such changes arise from ‘novel cognitive routines that control the sequence of cognitive processes required to perform the task’, and far-transfer effects only arise where the directly trained task yields a new cognitive routine that can be profitably applied to a far-transfer task. However, such routines will not benefit tasks for which well-established processes are used, amongst which Gathercole et al. [
60] include verbal short-term memory (STM) tasks that rely on phonological coding. This notion of ‘new cognitive routines’ receives some support from the overt strategies that some children in this study developed and used in the Odd One Out task, although the use of strategies was neither suggested nor encouraged by the researcher. One child used body parts, for example, the shoulders and head, to represent the location of left, right, and middle shapes, and performed the resulting sequence of gestures in visuospatial recall; another child tapped different fingers for different locations and used the resulting sequence of taps in recall. Strategies such as silent verbal rehearsal and chunking used in verbal recall tasks may be construed as ‘new cognitive routines’ that support recall capacity.
On Gathercole’s [
60] account, near-transfer effects to non-executive (STM) tasks would also be attributed to new cognitive routines induced by the training, rather than changes in children’s basic STM capacity. The implication is that increased auditory/visual attention and/or deployment of strategies induced by the training enabled children to use their basic STM capacity more effectively (rather than directly increasing this). On this interpretation, we might expect less benefit for a nonword repetition task, which requires immediate coding and production of a novel phonological form, potentially drawing on long-term knowledge of lexical phonology, but not amenable to strategies such as rehearsal or chunking [
61]. However, increases in nonword repetition have been reported following working-memory training in children with language difficulties [
37], indicating that this would be an interesting avenue to explore further.
What are the wider implications of the view that EWM interventions effect change in executive control of information processing that may benefit performance on language comprehension and also non-executive working-memory tasks? Consider everyday verbal inputs that children receive, for example, a sequence of verbal instructions from a teacher, or someone recounting an event or experience. Such inputs are rapid and transitory so must be processed and interpreted ‘online’. Comprehension therefore relies on immediate verbal STM and language processing, both drawing on established language knowledge. If the training has enabled children to deploy their STM capacity more effectively, as suggested above, this may extend the amount of verbal input they can process and recognise and thereby optimise the deployment of their language knowledge in online comprehension. If the training has enhanced their capacity to store and process information simultaneously, as suggested by gains on the executive working-memory tasks, this may support better retention of processed input while processing subsequent input and thereby improve their comprehension of stretches of verbal discourse. It should particularly benefit activities that do not involve rapid and transitory input, for example, reading comprehension and school-based learning that requires close attention to and the integration of new pieces of information.
Having reported the positive outcomes of this study, it is important to acknowledge its limitations. Most notably, due to resource limitations, the sample size was moderate, and the researcher who delivered the training also administered all the assessments so was not blind to the child’s group. This may have led to researcher bias, resulting in more favourable outcomes for the trained group. It is worth pointing out that outcomes in this group varied, and that some children showed little improvement. Nevertheless, replication of the study with larger samples and blinded assessment is needed to preclude possible bias. Further research is also needed to investigate whether there are wider benefits for language performance, of the sort proposed above, in the short and longer term.