Article
Reference
The effect of working memory training on a clinical marker of French-speaking children with developmental language disorder
STANFORD, Emily Nicole, DURRLEMANN, Stéphanie, DELAGE, Hélène
STANFORD, Emily Nicole, DURRLEMANN, Stéphanie, DELAGE, Hélène. The effect of working memory training on a clinical marker of French-speaking children with developmental language disorder. American Journal of Speech-Language Pathology , 2019, vol. 28, no. 4, p.
1388-1410
DOI : 10.1044/2019_AJSLP-18-0238
Available at:
http://archive-ouverte.unige.ch/unige:149969
Disclaimer: layout of this document may differ from the published version.
AJSLP
Research Article
The Effect of Working Memory Training on a Clinical Marker of French-Speaking Children
With Developmental Language Disorder
Emily Stanford,aStephanie Durrleman,aand Hélène Delagea
Purpose:Our work investigates the production of 3rd- person accusative clitic pronouns in French-speaking typically developing (TD) children and children with developmental language disorder (DLD) following a novel working memory (WM) training program (12 hrs of effective training) that specifically targets the components of WM that have been shown to be impaired in children with DLD and to be directly related to the mastery of clitics (Delage
& Frauenfelder, submitted for publication; Durrleman &
Delage, 2016).
Method:Sixteen TD children aged 5–12 years and 26 age- matched children with DLD completed our 8-week WM training program. Furthermore, an age-matched control group of 16 TD children and 17 children with DLD followed a scholastic training regime matched for intensity and frequency. Syntax and WM were assessed prior to and following the WM/scholastic training.
Results:Significant posttraining WM gains were found in TD children and children with DLD who took part in the WM training, and the production rate of 3rd-person accusative clitics significantly increased in children with DLD following the WM training. No significant WM or syntax gains were observed in the control group.
Conclusion:These findings are noteworthy as Melby- Lervåg and Hulme’s (2013) meta-analysis concluded that existing WM training programs show short-lived generalized effects to other comparable measures of WM, but that there is no evidence that such training generalizes to less directly related tasks. That our study led to gains in skills that were not trained (i.e., syntax) suggests that a WM training regime that is firmly grounded in theory and that targets the specific mechanisms shown to underpin the acquisition of syntax may indeed provide effective remediation for children with DLD.
O
ne of the hallmarks of developmental language disorder (DLD)1is impaired acquisition of expres- sive language, especially in the area of morpho- syntax, with nontarget-like utterances sometimes persisting into adulthood (Leonard, 2014). Cognitive development, however, is typically assumed to follow the expected trajec- tory in individuals with DLD, and a DLD diagnosis is often reached through a process of elimination, that is, no signs of neurological damage, no deficits in oral motor skills, no sensorial disabilities, and no pervasive developmental disorders. Thus,DLDis typically defined as a language im- pairment, with a clear dissociation between an impaired linguistic system and an intact general cognitive system.However, an increasing number of researchers have called
into question the idea of a deficit that is strictly limited to language, highlighting limitations in other domains that are frequently reported for individuals with DLD. Among these limitations is significant weakness on tasks that recruit verbal working memory (WM) resources, such as nonword repetition and digit span tasks. Results from various studies show that French-speaking children with DLD perform poorly on such tasks when compared to age-matched typi- cally developing (TD) children (Parisse & Mollier, 2008), as well as when compared to younger TD children matched for language ability (Majerus, Leclercq, et al., 2009). In light of this, certain researchers argue that DLD is in fact related to reduced general processing capacity, which then affects how parts of the grammar are acquired (Jakubowicz, 2004, 2005; Jakubowicz & Nash, 2001; Jakubowicz, Nash, Rigaut, & Gérard, 1998; Leonard, 2014). It is for the pre- viously noted reasons that we use the diagnostic terminol- ogy whose widespread adoption is supported by Bishop, Snowling, Thompson, Greenhalgh, and CATALISE
1Also known asspecific language impairment.
aFaculty of Psychology and Educational Sciences, University of Geneva, Switzerland
Correspondence to Emily Stanford: [email protected] Editor-in-Chief: Julie Barkmeier-Kraemer
Editor: Li Sheng Received October 2, 2018
Revision received February 11, 2019 Accepted April 15, 2019
https://doi.org/10.1044/2019_AJSLP-18-0238
Disclosure:The authors have declared that no competing interests existed at the time of publication.
Consortium (2017), as it reflects empirical findings that the difficulties associated with this population are not iso- lated to the domain of language but are more generally related to atypical cognitive development. On the basis of such evidence, which will be discussed in detail below, the focus of the current study was to investigate the effective- ness of cognitive training, in particular of WM, on the WM and syntactic capacities of children who have been diag- nosed with DLD and who demonstrate WM and syntactic deficits. We specifically wanted to test the hypothesis that improvements in WM would positively influence the use of third-person accusative clitic pronouns in French- speaking children with DLD. Such findings would be the- oretically significant, as they would further elucidate what is already known about the contribution of WM to lan- guage development, and they would be clinically relevant as they could open up new intervention perspectives for syntactic remediation.
DLD and Romance Clitics
Although a large number of authors have de- scribed various syntactic deficits in children with DLD (see Jakubowicz & Tuller, 2008; Parisse & Maillart, 2004, for findings on French-speaking children), the delayed acquisition of accusative clitic pronouns is considered a clinical marker of the disorder in French-speaking indi- viduals, as low production rates of such clitics reliably distinguish individuals with DLD from those without the disorder in French (Paradis et al., 2003; Tuller, Delage, Monjauze, Piller, & Barthez, 2011). Nominative and reflex- ive clitics, in contrast to accusative clitics, develop relatively unimpaired (Chillier-Zesiger, Arabatzi, Baranzini, Cronel- Ohayon, & Thierry, 2006). Moreover, low production rates of accusative clitics are primarily reported for those related to third person (3p), pointing not only to an asymmetry between different classes of clitics (e.g., nominative vs.
accusative) but also to a seeming asymmetry between different clitics within the same class, namely, first- and second-person accusative clitics versus 3p accusative clitics (henceforth 3p clitics; Delage, Durrleman, & Frauenfelder, 2016; Tsimpli, 2001; Tsimpli & Mastropavlou, 2007; Tuller et al., 2011).
Table 1 provides an abbreviated overview of French pro- nominal forms.
Researchers have identified various morphosyntactic features of accusative clitics that may contribute to their complexity and, by consequence, their delayed acquisition with respect to other clitic pronouns. Tuller et al. (2011) organize these features in terms of properties that apply to all accusative clitics (1) and (2) properties that are unique to 3p clitics (2).
(1) Properties of all accusative clitics
a. Noncanonical direct object position i. Jeanlevoit
1. John him sees 2. “John sees him”
b. Generally co-occur with a nominative clitic in spoken French
i. Jean,il levoit 1. John, he him sees 2. “John sees him” c. Nonlocally bound2
i. *Jeanileivoit 1. *Jeanihimisees 2. “John sees himself” (2) Properties specific to 3p clitics
a. Discourse independent
b. Morphologically marked for both number and gender
c. Unspecified for animacy
Essentially, the production of all accusative clitics in- volves movement from the thematic object position (the postverbal position) to a nonargument position (the prever- bal position), grouping with nominative clitics, and refer- ence to an antecedent not found in its immediate clause.
Furthermore, the production of 3p clitics requires a refer- ence that is established by the discursive context, as opposed to first person and second person whose reference is intrin- sically linked to the speech act. Finally, 3p clitics also in- volve agreement in both number and gender, but not in animacy, with the lack of specificity for animacy causing 3p clitics to be less semantically rich than other clitics. For the purposes of this article, we will consider two properties of 3p clitics that have been shown to contribute significantly to syntactic complexity (Durrleman & Delage, 2016), namely, the property of noncanonicity and that of morphological marking.
Syntactic Complexity
As 3p clitics present particular characteristics that render them complex, several proposals have been put for- ward in an attempt to define syntactic complexity. Compu- tational approaches, such as the derivational complexity hypothesis (DCH; Jakubowicz, 2005), describe complexity in terms of the number and nature of syntactic operations that have to be performed in order to arrive at the derived structure. Consequently, computationally complex struc- tures should be more difficult to process. According to Jakubowicz (2005, 2011) and Jakubowicz and Strik (2008),
Table 1.Abbreviated overview of French pronominal forms with the problematic 3p accusative clitic in bold.
Nominative Accusative Reflexive Dative Masc. Fem. Masc. Fem.
1p sg/pl je/nous me/nous me/nous me/nous
2p sg/pl tu/vous te/vous te/vous te/vous
3p sg/pl il/ils elle/elles le/les la/les se lui/leur Note. 3p = third person; Masc. = masculine; Fem. = feminine;
1p = first person; 2p = second person.
2The subject and the clitic cannot be coreferenced. Instead, the clitic must refer to an element not found in the same clause.
one syntactic operation that may contribute to computa- tional complexity is syntactic movement, or the displacement of a syntactic element from its base position to a position higher in the structure. According to the traditional trans- formational analysis (Kayne, 1975), the cliticization process involves movement of the object clitic from the postverbal argument position where it receives its thematic role to the preverbal surface position where it is spelled out (3).
(3) Jean le voit __
John him sees John sees him
Thus, not only do 3p clitics undergo syntactic move- ment, an operation that adds to the computational com- plexity of the derivation, but these elements also display a distinct noncanonical order that is well documented to be problematic for children with DLD (Marinis & Saddy, 2013;
Montgomery & Evans, 2009; van der Lely, 1996, for pas- sives; Adani, Forgiarini, Guasti, & van der Lely, 2014;
Contemori & Garraffa, 2010; Friedmann & Novogrodsky, 2004, for object relatives; Jakubowicz, 2011; Marinis & van der Lely, 2007; van der Lely, Jones, & Marshall, 2011, for object questions). As producing a 3p clitic involves the simultaneous retention of morphosyntactic information in memory and linking of this information to two different syntactic positions, the DCH can account for the delayed development of 3p clitics with respect to nominative and re- flexive. The DCH also predicts that less complex structures (e.g., canonical structures containing a lexical determiner phrase (DP) instead of a direct object clitic) will be prefer- entially produced, even in infelicitous contexts, until suf- ficient cognitive resources to compute the more complex structure become available.3
It is worth briefly noting that another approach, set within a phonological/prosodic framework, also offers an interesting perspective on noncanonicity and how this re- lates to the complexity of 3p clitics. Demuth (1992, 1994, 2015) proposes that prosodic constraints on children’s early productions could be responsible for the inconsistent use of certain grammatical morphemes in young children.
This account hinges on the prosodic hierarchy developed by Nespor and Vogels (1986, 2007), which defines vari- ous prosodic categories, one of which is the syllable, a class that can be divided into strong (stressed) or weak (un- stressed) syllables. Gerken (1991, 1994, 1996) illustrates that young TD children show a strong preference for strong– weak syllable sequences and are more likely to omit a weak syllable in an utterance when the opposite sequence occurs. This is corroborated by cross-linguistic research that reports that Italian-speaking children with DLD are far more likely to omit 3p clitics than age-matched English-speaking
children with DLD are to omit object pronouns despite the fact that both types of pronouns are weak (Leonard, 2014;
Leonard, Sabbadini, Volterra, & Leonard, 1988). The criti- cal difference between the two languages is the position in which the object pronoun occurs in the phonological phrase: preverbally in Italian, as in French, and post- verbally in English. As a result of its noncanonical posi- tion, the Romance object clitic requires the production of a weak–strong sequence (4), whereas the canonical En- glish pronoun can occur in a strong–weak pattern (5).
Thus, it is possible that the interaction between syntax and prosody further increases the demands associated with producing 3p clitics as one must not only retrieve grammatical information but also integrate this infor- mation into the prosodic organization in the form of an unstressed syllable.
(4) [la mère] [levoit]
(5) [the mom] [seeshim]
WM
Working memoryrefers to the cognitive activity of temporarily holding and, if necessary, transforming infor- mation during an ongoing mental task. According to the most influential WM model in psycholinguistics, Baddeley and Hitch’s (1974) multicomponent model, the central exec- utive, a limited-resource attentional control system, coordi- nates two slave systems: the phonological loop, which stores verbal information, and the visuospatial sketchpad, which stores visuospatial information. Additionally, a fourth com- ponent, the episodic buffer, was added to the model and is described as a functional interface between the slave sys- tems and long-term memory (Baddeley, 2000). One advan- tage of Baddeley’s multicomponent model is that it allows simple cognitive activities, such as storing verbal items in the phonological loop (a simple span task), to be dissoci- ated from more complex cognitive activities that require the important involvement of the supervisory central exec- utive system, such as storing verbal items while simulta- neously performing an additional processing task (a complex span task). It has been suggested that the most accurate way to assess phonological loop capacity is through simple span tasks, such as the repetition of nonwords or the forward repetition of digits, and that the central executive capac- ity is best measured through complex span tasks, such as the backward repetition of digits or listening span tasks (Barrouillet & Camos, 2007).
WM and DLD
Substantial deficits in verbal WM have been reported in the literature for children with DLD, for both simple span tasks (Archibald & Gathercole, 2006; Delage & Frauenfelder, submitted for publication; Majerus & van der Linden, 2003;
Montgomery & Evans, 2009) and complex span tasks (Archibald & Gathercole, 2006; Delage & Frauenfelder, submitted for publication; Hoffman & Gillam, 2004;
Montgomery, 2000). Thus, it has been concluded that the
3Although it will not be discussed here, it should be noted that another computational account has been put forward to describe the delayed development of 3p clitics, which argues that their complexity stems from the crossing of movement chains created during the derivation of the structure. See Zesiger et al. (2010) for a full review of this account.
WM deficit found in children with DLD occurs at both the level of the phonological loop and the level of the central executive. It is worth noting that many studies report that visual WM, associated with the visuospatial sketchpad in Baddeley’s model, is relatively preserved in children with DLD (Archibald & Gathercole, 2006; Archibald & Joanisse, 2013; Baird, Dworzynski, Slonims, & Simonoff, 2010).
Within the phonological loop, Majerus, Poncelet, Greffe, and van der Linden (2006) distinguish item and serial order short-term memory (STM), the former concern- ing the retention of the phonological and semantic proper- ties of verbal stimuli and the latter concerning the short-term storage and retrieval of the successive order of a series of verbal stimuli. To investigate the precise nature of the pho- nological loop deficit in children with DLD, Majerus, Leclercq, et al. (2009) tested separately item and serial order STM capacities of 12 children with DLD aged 6–10 years. These children presented difficulties in the serial order reconstruction task when compared to age-matched TD controls, results that were corroborated by Delage and Frauenfelder (submitted for publication) when they tested serial order STM in 28 children with DLD aged 5–14 years. Furthermore, Delage and Frauenfelder found that DLD performance in serial order reconstruction pre- dicted a significant part of both mean length of utterance in spontaneous speech and rate of embedding in repetition of complex structures and in spontaneous speech.
With regard to the central executive in children with DLD, one of the most widely employed tasks used to as- sess its capacity is the listening span task, in which children are asked to listen to a set of sentences and retain the last word of each sentence for subsequent recall. To ensure that the children listen to the whole sentence, they must make a true–false judgment about each sentence while simultaneously retaining the final word. Studies show that children with DLD often perform below age-matched peers on such tasks (Archibald & Gathercole, 2006; Weismer, Evans, & Hesketh, 1999) and that performance is influenced by a number of factors, such as the frequency of the words to be recalled, the number of lexical competitors associated with the words, and the precision of the children’s semantic knowledge of the words (Mainela-Arnold & Evans, 2005; Mainela-Arnold, Evans, & Coady, 2010). Other central executive assessments, such as the processing of visual or auditory stimuli, focus more specifically on resource allocation or attention, rather than the storage and mental manipulation of verbal informa- tion. While the literature on attention in DLD is inconsistent, there is nonetheless strong evidence that children with DLD present difficulties in sustained attention (Ebert & Kohnert, 2011; Finneran, Francis, & Leonard, 2009).
WM and Syntax
Many studies have shown a direct link between WM and lexical acquisition (Gathercole, Service, Hitch, Adams,
& Martin, 1999; Leclercq & Majerus, 2010; Majerus et al., 2006), and more recent research has highlighted the rela- tionship between WM and syntactic development. Research
on TD children has reported that weak simple span capacity negatively affects the number of complex sentences repeated by 4- to 5-year-old children (Willis & Gathercole, 2001), while research on 6- to 8-year-old children with DLD has demon- strated that complex span rather than simple span is asso- ciated with the repetition of complex syntactic structures, such as relative clauses (Frizelle & Fletcher, 2015). Addi- tionally, using regression analyses, Delage and Frauenfelder (in press) reported that, in 48 TD French-speaking children aged 6–12 years, simple span measurements predict perfor- mance on tasks that require the participant to repeat com- plex structures and that complex span measurements predict performance (e.g., mean length of utterance or rate of deep embedding) when spontaneous speech is analyzed. When the study was replicated with 28 children with DLD, re- gression analyses revealed that simple span, along with complex span and age, explained 58% of the variance in the repetition of complex sentences, and that simple span was an important predictor of complex syntax in spontaneous speech (Delage & Frauenfelder, submitted for publication).
In comprehension, Montgomery, Magimairaj, and O’Malley (2008) observed a significant positive correlation between complex span and the comprehension of copular passives and of structures containing various binding prin- ciples in 6- to 12-year-old TD children, findings that were then replicated for children with DLD (Montgomery &
Evans, 2009). Similarly, Marinis and Saddy (2013) found a relationship between performance on passives and com- plex span in children with DLD. Furthermore, regression analyses revealed that complex span significantly predicts the comprehension of complex structures in school-age TD children (Delage & Frauenfelder, in press) while simple span, complex span, and age explain a significant part of the variance in the comprehension of complex structures in children with DLD (Delage & Frauenfelder, submitted for publication).
With regard to the specific relationship between WM and 3p clitics, correlations have been found cross- linguistically in both typical and atypical populations, in both first- and second-language learners, and for both pro- duction and comprehension. Using a nonword repetition task, Mateu (2015) found a correlation between simple span and the ability to accurately interpret 3p clitic con- structions in 32 TD Spanish-speaking children aged 2–4 years.
In production, Grüter and Crago (2012) found that lower scores on nonword repetition and backward digit recall were strong predictors of how frequently 3p clitics were omitted and higher scores predicted frequency of object clitic production by 32 school-age L2 French children.
Finally, Durrleman and Delage (2016) found that complex span (measured through backward digit recall) correlated with the number of correctly produced 3p clitics in forty- three 5- to 16-year-old French-speaking children displaying atypical development (autism spectrum disorder,n= 21 and DLD,n= 22).
The evidence presented above clearly shows a strong relationship between verbal WM and complex syntax, which
has potential clinical implications. If WM limitations do indeed predict complex syntax deficits in children with DLD, then it could be promising to train WM in this popu- lation in an attempt to remediate such deficits.
WM Training
As WM impairments in populations other than DLD are increasingly unveiled (Durrleman & Delage, 2016;
Williams, Goldstein, & Minshew, 2006, for autism; Goodglass, Gleason, & Hyde, 1970; Gordon, 1983, for aphasia; Brady, 1991; Elbro, 1996; Wagner & Torgesen, 1987, for dyslexia;
Dowson et al., 2004; Karatekin, 2004; Karatekin & Asarnow, 1998; Kuntsi, Oosterlaan, & Stevenson, 2001; Westerberg, Hirvikoski, Forssberg, & Klingberg, 2004, for attention- deficit/hyperactivity disorder [ADHD]) and as the important contribution of WM to cognitive development is better under- stood (Case, 1985; G. Cohen & Conway, 2007; Gathercole
& Alloway, 2006, 2008), a number of WM training pro- grams have been created and commercialized with the aim of enhancing WM capacity. The idea is that, if WM is a mental workspace that supports a wide range of cognitive tasks, training WM should lead to transfer effects to such tasks, even if they were not the specific target of the training regime. Activities that train WM generally require one to encode, maintain, or manipulate information, which can be verbal or visual in nature (Borella, Carretti, Riboldi, &
De Beni, 2010). For example, a verbal WM training activ- ity might involve the forward and backward recall of a series of digits, and such an activity would be paired with feed- back of the individual’s performance as a motivational tool (Beck, Hanson, Puffenberger, Benninger, & Benninger, 2010). Furthermore, WM training programs are typically adaptive, an activity either increases or decreases in diffi- culty according to the individual’s performance, in order to maximize potential gains (Klingberg, Forssberg, &
Westerberg, 2002).
CogMed WM training (http://www.cogmed.com/) is one of the most well-known and widely used computer-based memory training programs currently available on the mar- ket. This program offers a variety of activities that are ad- vertised to train visuospatial and verbal WM capacity and consequently to significantly improve performance in un- trained cognitive tasks that are supported by WM. Several studies have investigated the accuracy of such claims in an attempt to analyze the effectiveness of CogMed and other commercial WM training programs, such as Jungle Memory (http://www.junglememory.com/) and Cognifit (http://www.
cognifit.com/), but the conclusions are often contradictory, and the studies have been criticized for their methodologi- cal shortcomings (Shipstead, Redick, & Engie, 2010). In particular, Melby-Lervåg and Hulme (2013) outline three key points that should be taken into consideration when conducting a WM training study: (a) participants should be randomly assigned to the different groups, (b) the perfor- mance of the trained group must be compared to that of a suitable control group, and (c) an alternative training program should be compared to that of the WM training program.
In their meta-analysis of studies that met these criteria, Melby- Lervåg and Hulme conclude that, for TD children and healthy adults, existing WM training programs show short-lived generalized effects to other comparable measures of WM, but that there was no evidence that such training transfers to less directly related tasks. As for children with language difficulties, Holmes et al. (2015) reported some WM improvements after 8 weeks of CogMed training, but these improvements were only significant for visuospatial WM.
However, the findings from this study should be interpreted with caution as the study did not respect the methodologi- cal criteria defined by Melby-Lervåg and Hulme.
According to Majerus (2016), another potential rea- son for the absence of more convincing results from WM training studies, along with methodological flaws, is that WM training programs such as CogMed are overly general and do not target the specific weaknesses documented for a particular population. For example, the majority of the WM activities proposed by CogMed focus on visuospatial WM capacity. However, as previously mentioned, children with DLD have known verbal WM deficits, and verbal WM capacity predicts language competence, with most studies reporting that visual WM is spared in the DLD population.
Thus, it seems reasonable to suggest that a WM training program designed to pinpoint the specific WM components shown to be related to the development of particular linguis- tic elements in DLD would be more appropriate for this population. For this reason, we created a novel WM training program, Magic Memory, which we hope is better adapted to children with DLD and to the WM components on which we wish to focus. For example, in our program, we do not include visuospatial WM activities but rather concentrate on the previously identified pockets of WM weakness found in DLD. To date, WM in children with DLD is characterized by deficits in both simple and complex verbal spans, and we turned in particular to studies that show clear links between such spans and syntax when designing our activities. Two types of tasks are of special interest, serial STM (a simple span task) and counting span (a complex span task), as previous research shows that these tasks predict the larger part of variance of syntactic complexity in TD children and in children with DLD (Delage & Frauenfelder, in press, submitted for publication). Interestingly, these two tasks are less language dependent than other tasks of verbal WM, such as nonword repetition or listening span, since mini- mal phonological encoding is required to complete them.
We thus train verbal WM by means of simple and complex span tasks, with a maximal limitation of language process- ing. The rationale for training WM is that, if verbal WM deficits underlie some of the syntactic difficulties of children with DLD, then intensively exercising the brain networks in- volved in verbal WM tasks to improve their efficiency may also lead to ameliorations in syntax.
Aims
The main aim of the current study was to investi- gate the effectiveness of WM training on the WM and
syntactic capacities of children who have been diagnosed with DLD and who demonstrate WM and syntactic defi- cits. More precisely, we wanted (a) to determine whether our novel, targeted WM training program, Magic Mem- ory, improves the efficiency of the WM components that have been trained in this population (a near-transfer effect) and (b) to investigate the possibility that an improved WM system leads to improvements in domains that were not directly trained but that are arguably reliant on WM resources, such as syntax (a far-transfer effect). We chose not to use an existing commercial program but to design our own so that we could adapt the training to our clini- cal population and focus specifically on the WM compo- nents we wished to train. For example, we did not include visuospatial WM activities in our program, although such activities are frequently present in more general WM programs.
We were specifically interested in the potential in- fluence that a targeted WM training could have on the production of 3p clitics. As previously underlined, delay in mastering 3p clitics is a clinical marker for DLD in French-speaking children (Tuller et al., 2011) and may stem from this element’s computational complexity relat- ing to WM maturation, which is also delayed in DLD.
Our work thus investigates the production of clitic pro- nouns in French-speaking children with DLD prior to and following our 8-week WM training program, Magic Memory. This program was designed to specifically fo- cus on the components of WM that are impaired in the DLD population (Delage & Frauenfelder, submitted for publication) and are directly related to the mastery of 3p clitics (Durrleman & Delage, 2016).
We base our work on accounts of DLD that main- tain that reduced processing capacity affects performance on high cognitive load syntax tasks in children with DLD (Jakubowicz, 2011), a claim that is supported by empirical evidence (Delage, 2015; Delage & Frauenfelder, submitted for publication). We hypothesize that training the WM capacities of children with DLD will not only lead to in- creased capacity in the WM components that have been trained but will also result in observable knock-on effects in the domain of syntax. More specifically, for children who follow the WM training regime, we expect posttrain- ing WM scores to be higher than the pretraining scores.
In this group, we also expect to observe a significant in- crease in the number of 3p clitics produced following the WM training, and this increase should be related to the improvement in WM.
Finally, as this is the first study to test our novel WM training program Magic Memory, TD children were in- cluded in order to obtain a more complete picture of the effectiveness of this particular training program. While TD children are not expected to display deficits in WM or in syntax, a predictive relationship between WM and syntax has been reported for this population (Delage & Frauenfelder, in press). Thus, we hypothesize that, for TD children whose WM and linguistic systems are still developing, Magic Memory may have a beneficial effect.
Method
ParticipantsSeventy-five children from Geneva, Switzerland, and Paris, France,4participated in this study. All children were either monolingual (n= 62,M= 8;8 [years;months],SD= 1;11) or early bilinguals who had acquired French before the age of 3 years (n= 13,M= 9;1,SD= 2;0). One bilin- gual participant was excluded from the study because she was receiving a formal bilingual education; all other partic- ipants, monolingual and bilingual, attended monolingual French schools. Finally, all of the children included in this study were within the normal range (≥10th percentile) for nonverbal reasoning (Raven’s Progressive Matrices; Raven, Court, & Raven, 1998).
Participants with DLD included 43 children aged 6;0–12;5 (M= 8;11,SD= 1;9), an age range for which a pre- dictive relationship between WM and syntax has been shown for both TD children and children with DLD (Delage &
Frauenfelder, in press, submitted for publication). All of the children in this group received regular speech-language services (one to two sessions per week) and were recruited by postgraduate students who were in direct contact with the relevant clinicians. Each participant had been diagnosed with DLD by a qualified specialist, and syntactic deficits had been identified at the time of diagnosis. We further assessed expressive syntax in French via the standardized test Bilan Informatisé de Langage Oral (BILO-3C; Khomsi, Khomsi, Pasquet, & Parbeau-Guéno, 2007) and impairment in this realm, which was a requirement for our clinical popu- lation, was confirmed for all participants with DLD by a score of at least 1.25SDs below age-specific norms. Ad- ditionally, we used the French standardized Evaluation of Working Memory Test (Boutard & Gatignol, 2015) to evalu- ate phonological loop (simple span) and central executive (complex span) capacity. Each participant with DLD dem- onstrated impaired performance (at least 1.25SDs below the normative mean) on a minimum of three of the six WM tasks, thus meeting the inclusion requirements for this group.
The results of independentttests revealed that there was no significant difference between the 32 monolingual and 11 bilingual participants for syntax,t(41) = 1.90,p= .07.
There was, however, a significant difference between the two groups for simple span,t(41) = 2.73,p= .01,d= 0.70, and for complex span,t(41) = 2.34,p= .02,d= 0.53, with the bi- lingual participants showing less impaired simple and com- plex span capacity than the monolingual participants.
However, as the two groups performed comparably on the measure of syntax, the results of the monolingual and bilingual children will be presented together.
Thirty-two TD children aged 5;2–12;7 (M= 8;5, SD= 2;1) were also included in this study. Children in
4The differences between Swiss French and standard French are minor and mostly limited to a small number of lexical variations. Name agreement for all test items used in our study was verified by both a native French speaker from Switzerland and a native French speaker from France prior to the start of the study.
this group were screened to ensure they had typical syntax (≥ −1 SDon the BILO-3C) and WM (no more than two of six scores that were≤ −1SDbelow the normative mean on the Evaluation of Working Memory Test). Independent ttests revealed that, even though both the TD children and children with DLD were in the normal range for nonverbal reasoning, the performance of TD children was significantly better than that of children with DLD,t(73) = 6.18,p< .001, d =1.04. We considered this unproblematic as findings from prior research suggest that children with DLD typi- cally perform below age-matched TD peers on nonverbal tasks (Leonard, 2014). Furthermore, the definition of DLD as described by Bishop et al. (2017) specifies that nonverbal delays are no longer exclusionary for a DLD diagnosis, provided the delays are not related to intellectual disability.
Independentttests also revealed that TD children signifi- cantly differed from children with DLD on the standardized tasks of syntax and WM: TD children performed significantly better than children with DLD on syntax,t(73)=11.21, p< .001,d= 1.90; simple span tasks,t(73) = 10.54,p< .001, d= 1.79; and complex span tasks,t(73) = 8.48,p< .001,d= 1.37. These results were expected as one inclusion criterion of our study stated that participants with DLD were re- quired to have impaired syntax and WM, whereas TD participants were required to meet normative standards for syntax and WM. Of the TD participants, 30 were mono- lingual and two were bilingual but had been exposed to French before the age of 3 years. Again, monolingual and bilingual results will be presented together for the TD group. Approval for this study was obtained from the Ethics Committee of the Faculty of Psychology and Educational Sciences at the University of Geneva, and parents of all participants gave informed, written consent for their chil- dren to participate in this research.
Procedure
In a one-to-one setting, participants completed a set of pretraining tests in syntax and in WM that lasted approxi- mately 1.5–2 hr. These assessments were administered by graduate students in speech-language pathology programs in Geneva, Paris, and Tours roughly 1 week prior to the commencement of the training program and took place either at the home of the child or, in the case of some partici- pants with DLD, in a consulting room at a private speech therapy practice. Just over half of the participants (n= 42;
26 DLD, 16 TD) then took part in an 8-week WM training program that consisted of three 30-min sessions per week, while the remaining participants (n= 33; 17 DLD, 16 TD) were placed in a scholastic training group that followed the same training rhythm and intensity as that of the WM train- ing group (see Table 2). A scholastic training group was included to respect the methodological criteria outlined by Melby-Lervåg and Hulme (2013) and to ensure that any potential gains in WM and/or syntax following the training period were indeed related specifically to the WM train- ing program rather than maturation or global cognitive stimulation. Assignment to the different training groups was
random, although the WM training commenced roughly 1 month before the scholastic training. For this reason, the first participants to join the study were necessarily assigned to the WM training group. Independentttests showed that children with DLD in the two training groups did not sig- nificantly differ in measures of age, syntax, WM (neither simple nor complex span), or nonverbal reasoning. This was also true for TD children (see Table 3). Similar analy- ses confirmed that there were no significant differences be- tween the TD children in the WM training group and those in the scholastic training group. All training took place at the child’s home or, in the case of a few children with DLD, at the child’s speech-language pathologist’s clinic under the supervision of a parent and/or a graduate student. Training sessions that took place in a clinic did not replace the regu- lar speech-language services these children received, and parents were not charged for the time spent at the clinic during these sessions. When the student was unable to be present at one of the sessions, he or she was in contact with the family in order to verify that the session had been suc- cessfully completed. Due to software bugs at the beginning stages of our study, nine of the 26 participants with DLD in the WM training group did not manage to complete the full 24 training sessions, but all 26 of these participants did successfully complete at least 20 of the 24 sessions. All 16 of the TD participants in the WM training group and all children, both DLD and TD, in the scholastic training group completed all 24 training sessions. Upon completion of training, a posttraining assessment in syntax and in WM was administered to the participants in individual sessions using a parallel form of the pretraining assessment. In other words, the measurement characteristics of the pre- and posttraining tests were identical, but the two versions con- tained different items that had been matched for difficulty.
The posttests were administered approximately 1 week after the final training session.
Pre- and Posttraining Tests: WM and Syntax Before and after training, participants completed six WM tests: three measuring verbal STM (i.e., simple span tasks) and three measuring verbal STM and executive con- trol (i.e., complex span tasks). The three simple span tasks included forward digit recall (Wechsler Intelligence Scale for Children–Fourth Edition; Wechsler, 2005), serial order word span (Majerus et al., 2006), and nonword repetition (Batterie d’Évaluation du Langage Écrit [BELEC]; Mousty, Laybaert, Alegria, Content, & Morais, 1994), and the three complex span tasks included backward digit recall (Wechs- ler Intelligence Scale for Children–Fourth Edition; Wechsler, 2005), counting span (Case, Kurland, & Goldberg, 1982), and running span (Pross, Gaonac’h, & Gaux, 2008). Stan- dard scores were derived for the individual tests, and com- posite scores were calculated by averaging standard scores for each set of tests (simple and complex span). In addition to the WM tests, participants also completed a battery of pre- and posttraining syntax tests measuring receptive and expressive syntax. However, for the purposes of this
article, the only syntactic structure that will be discussed is one that has been identified as a clinical marker of DLD, namely, clitic pronouns. Independentttests con- firmed that there was a significant difference between our two cognitive groups on all pretest measures (TD WM/
scholastic children outperformed DLD WM/scholastic chil- dren) but that our two training groups contained partici- pants with similar pretest abilities (DLD/TD WM children did not differ from DLD/TD scholastic children). Descrip- tive statistics for the pretests are provided in Table 4.
Clitics Task
To test the production of clitic pronouns, we used a Production Probe for Pronoun Clitics task (Delage et al., 2016; Tuller, Audollent, Delage, & Monjauze, 2004) that required participants to respond to a question about an im- age that appeared on a computer screen. The task, which elicited 3p nominative and accusative pronouns, contained three training items, 12 test items, and four distractors (see Table 5). Both animate and inanimate arguments were used as referents for the 3p clitics, and items contained ac- cusative and the nominative pronouns. Sample items are provided in (6) and (7), along with the corresponding im- ages (see Figures 1 and 2). All testing sessions were audio- recorded and were subsequently transcribed, coded, and checked twice by two different experts.
(6) Experimenter: Que fait Marie avec le fil du téléphone?
“what does Mary with the wire of the telephone
“What is Mary doing to the telephone wire?”
Expected response: Elle le coupe.
she him cuts
“She’s cutting it.” (7) Experimenter: Que fait Marie avec la voiture?
what does Mary with the car
“What is Mary doing to the car?” Expected response: Elle la regarde.
she her looks
“She’s looking at it.”
WM Training
Participants completed 24 sessions of WM training using the application Magic Memory that was specifically created for this study by our team at the University of
Geneva. Magic Memory is an interactive 8-week training pro- gram that consists of five original activities that train both simple and complex spans. The five activities, described be- low, were explicitly designed to target previously identified WM weaknesses in children with DLD and to train the com- ponents of WM that have been shown to be the most predic- tive of performance on tests assessing complex syntax in TD children and in children with DLD (Delage & Frauenfelder, in press, submitted for publication). The training took place in the form of an iPad5application conceived and created by a research team at the University of Geneva. In order to maximize the potential positive impact of the training pro- gram on WM and syntactic capacities, each participant completed three training sessions per week, with one session lasting approximately 30 min (±5 min per activity) and the total amount of training time adding up to 12 hr per par- ticipant. For each activity, task difficulty was adapted to match the child’s performance level on a trial-by-trial basis.
Furthermore, the order of the activities was randomized for each session to prevent order (or fatigue) effects in the data.
The participants were provided with positive feedback at the end of each activity and at the end of each session in order to boost motivation and to encourage active engage- ment in each training session.
Magic Memory Activities Activity 1: Serial Order Memory
In order to train serial order STM capacity, the first activity, inspired by Majerus et al.’s (2006) serial order recon- struction task, required participants to retain the successive order of a series of familiar monosyllabic verbal items.
After familiarizing the child with the lexicon of the task, which was made up of eight common nouns that had been controlled for name agreement, age of acquisition, and lexi- cal frequency, the child listened to a series of items from the task lexicon, such asfeu(traffic light),banc(bench), andarbre(tree). The items were randomly selected from the pool of eight nouns in the task lexicon, and no item occurred twice in the same list. Next, images corresponding to the se- ries of orally presented verbal items appeared on the screen, along with a train containing empty wagons, the number of which matched the number of heard items (see Figure 3).
5A Windows version of Magic Memory was also created so that families without an iPad were able to participate in the study. Of the Magic Memory participants, 37 completed the Magic Memory training using an iPad, whereas five used the Windows version on a computer.
Table 2.Summary of participant information according to training group.
Training group Cognitive group n Sex Age range (years;months) Age,M(SD) (years;months)
WM training DLD 26 7F, 19M 6;5–12;5 8;8 (1;7)
TD 16 12F, 4M 5;2–12;4 8;8 (2;0)
Scholastic training DLD 17 6F, 11M 6;0–12;2 9;1 (1;5)
TD 16 9F, 7M 5;4–12;7 8;1 (1;8)
Note. WM = working memory; DLD = developmental language disorder; F = female; M = male; TD = typically developing.
In other words, if the child heardfeu, banc, andarbre, images representing these three items would appear on the screen, along with a train containing three empty wagons.
The child was then asked to place the images in the correct order in the wagons, that is, in the order in which they had been presented. Because the images of the items were available to the participant during the recall phase, only the order of presentation had to be retained and retrieved.
As the participant improved, the number of presented items increased, thus increasing the complexity of the task.
Activity 2: WM Updating
With the second activity, we aimed to train WM by creating an adapted version of the classicn-back task. This activity involved dual processing as it required the participants to store one to three visual elements in WM (depending on
the participant’s level) and to frequently update the contents of WM. During this activity, the participants were presented with a sequence of visual stimuli that had been controlled for visual complexity, lexical frequency, and age of acquisi- tion and were then instructed to indicate when the current stimulus matched the one from one, two, or three steps ear- lier in the sequence according to the level of difficulty. The load number adapted to the ability of the participant to make the activity more or less challenging. When the par- ticipant achieved a score greater than 71% at the end of a session, then-back level increased by one for the next ses- sion. For scores of 51%–71%, then-back level stayed the same from one session to the next, and for scores below or equal to 50%, then-back level decreased by one for the next session. During each session, a visual reminder was present on the screen to inform participants of the type
Table 4.Summary of composite working memory (WM) scores and syntax scores on pretests.
Participant
Simple span Complex span Clitics
M(%)
(SD) Comparison M(%)
(SD) Comparison M(%)
(SD) Comparison
DLD 37.58
(9.49)
*** 22.73
(7.83)
*** 33.33
(28.23)
***
TD 56.03
(11.88)
41.67 (15.04)
82.55 (21.62)
DLD, WM 38.09
(9.82)
ns 23.98
(7.97)
ns 32.69
(28.96)
ns DLD, scholastic 36.81
(9.20)
20.81 (7.47)
34.31 (27.93)
TD, WM 57.08
(13.66)
ns 42.24
(13.19)
ns 79.17
(22.15)
ns TD, scholastic 54.98
(10.15)
41.09 (17.12)
85.94 (21.24) Note. DLD = developmental language disorder; TD = typical development;ns= not significant.
***p≤.001.
Table 3.Summary ofzscores on standardized tests.
Participant
BILO-3C Simple span Complex span Matrix reasoning
M
(SD) Comparison M
(SD) Comparison M
(SD) Comparison M
(SD) Comparison
DLD −2.90
(1.60)
*** −1.43 (1.10)
*** −2.23 (0.75)
*** −0.41
(0.78)
***
TD 0.82
(1.13)
0.95
(0.75) −0.62
(0.90)
0.69 (0.72)
DLD, WM −2.61
(1.44)
ns −1.18
(1.16)
ns −2.11
(0.82)
ns −0.32
(0.81)
ns DLD, scholastic −3.35
(1.77) −1.81
(0.90) −2.42
(0.60) −0.54
(0.75)
TD, WM 0.71
(1.12)
ns 0.93
(0.82)
ns −0.64
(0.80)
ns 0.49
(0.73)
ns TD, scholastic 0.93
(1.17)
0.97
(0.70) −0.60
(1.01)
0.88 (0.67)
Note. BILO-3C = Bilan Informatisé de Langage Oral; DLD = developmental language disorder; TD = typical development; WM = working memory;ns= not significant.
***p≤.001.
of cognitive task to be performed: 1-, 2-, or 3-back. Each session consisted of five 1-min blocks, and each block contained 20 trials. Participants were given approximately 3 s to respond during each trial and received either positive or negative feedback after each response. The visual items presented in each block/session were divided as follows:
two animals (e.g., an elephant), one vegetable or fruit (e.g., a carrot), and three everyday objects (e.g., a book).
Activity 3: Serial Order and Complex WM
The third activity, inspired by Turner and Engle’s (1989) operation span task, was designed to train complex span and required participants to retain the order of famil- iar auditory stimuli (storage phase) while simultaneously performing a secondary task, a visual discrimination task (processing phase). After familiarizing the participants with the task stimuli, which were made up of eight common environmental sounds that had been controlled for familiarity and representativeness, the child listened to a series of items from the set of task stimuli, such as a ringing telephone, a crying baby, and a barking dog. The items were randomly selected from the pool of eight sounds in the task stimuli, and no item occurred twice in the same list. Immediately following the presentation of each item to be stored, partici- pants were asked to complete a rapid visual comparison of
quantity task. This processing task presented the partici- pant with two side-by-side pages containing a number of everyday items and asked the participant to rapidly deter- mine which page contained the most items. Each processing trial contained three consecutive items. After having listened to a certain number of stimuli and after having completed the secondary task following each stimulus (see Figure 4), the participant was presented with images corresponding to the sounds and asked to place them in the correct order, that is, in the order in which they were presented. The number of presented stimuli in each trial increased or decreased successively based on the child’s performance.
Activities 4 and 5: Simple and Complex Span
The fourth and fifth activities aimed to train simple and complex WM span. They were inspired by the classic digit span task, but rather than ask participants to retain the order of a series of digits, they asked participants to retain the order of a series of color names (e.g., blue, red, green). For both activities, the participants were familiar- ized with the task lexicon, which consisted of eight mono- syllabic color names controlled for frequency, at the beginning of each session. For the fourth activity, which trained simple span, the participants heard a sequence of two to eight colors (sequence length being determined by performance) and were then presented with images corre- sponding to all eight colors in the task lexicon. The par- ticipants were instructed to select the colors they had heard in the order of presentation. More precisely, the participants needed to retain both item information (color name) and serial order information (the order of presen- tation) in order to successfully complete this activity. The procedure for the fifth activity was identical to that of
Table 5.Clitic pronouns elicited by clitics task.
Gender Nominative Accusative
3p Masculine il (6) le (5)
Feminine elle (6) la (5)
l’(2)a Note. 3p = third person.
aThis is the French definite article that is used before a singular noun (masculine or feminine) that begins with a vowel or a muteh.
Figure 1.Illustration corresponding to the example item in (6).
Reprinted fromLingua, Vol. 121, Tuller, Delage, Monjauze, Piller, and Barthez,“Clitic Pronoun Production as a Measure of Atypical Language Development in French,”pp. 423–441, Copyright © 2011, with permission from Elsevier.
Figure 2.Illustration corresponding to the example item in (7).
Reprinted fromLingua, Vol. 121, Tuller, Delage, Monjauze, Piller, and Barthez,“Clitic Pronoun Production as a Measure of Atypical Language Development in French,”pp. 423–441, Copyright © 2011, with permission from Elsevier.
the fourth activity, but this time the participants were asked to recall the color names in reverse order, a complex span task. As with the other activities, the level of difficulty was adjusted to match the ability of the participant. For example, after two consecutive correct responses with a sequence length of three, the next list of items would have a sequence length of four. Likewise, after two consecutive incorrect responses with a sequence length of three, the next list of items would have a sequence length of two.
Scholastic Training
Participants in the scholastic training group com- pleted 24 training sessions using the online learning plat- form Squla (http://www.squla.fr). During their sessions, these participants were able to choose from several edu- cational games that revised a variety of subjects, such as spelling, history, and geography, but which did not specifi- cally target the development of WM capacity. Session dura- tion and training frequency was matched to that of Magic Memory (three 30-min sessions per week for 8 weeks). Regard- ing multimedia support, participants were given the choice between using a tablet or a computer (nTABLET= 4;nPC= 29).6As with Magic Memory, Squla offered an adaptive training program that provided the participants with frequent feedback.
Results
Before investigating if the performance of children who had taken part in the WM training had improved on
WM tasks that had not been trained, we wanted to confirm that performance had improved on the WM tasks that had been trained, namely, the Magic Memory activities.
To calculate this, we took the maximum span mean of the first four sessions and compared it to the maximum span mean of the final four sessions for four of the five activities.7For children with DLD, the results of indepen- dentttests revealed that there was a significant difference between the two means for all four activities, with max span being larger at the end of training than it was at the beginning:t(6) = 7.60,p< .001,d= 3.74 for Activity 1;
t(6) = 3.81,p< .01,d= 1.89 for Activity 3;t(6) = 3.90, p< .01,d= 1.93 for Activity 4;t(6) = 6.17,p< .001,d= 3.08 for Activity 5. Figure 5 charts the performance trend for participants with DLD,8with an average span increase of 1.5 items from the beginning to the end of the training for the four activities.
For TD children, there was also a significant max span increase from the first four sessions to the final four sessions for all four activities:t(6) = 4.53,p= .01,d= 2.26 for Activity 1;t(6) = 4.46,p= .01,d= 2.24 for Activity 3;
t(6) = 5.66,p< .01,d= 2.82 for Activity 4;t(6) = 5.22, p< .01,d= 2.59 for Activity 5. See Figure 6 for the charted performance trend for the TD participants.9
Next, to investigate whether the individual training programs led to significant changes in performance on the WM tasks that had not been trained, repeated-measures analyses of variance (ANOVAs) were run with time (pre- test, posttest) as the within-subject factor and training type (WM, scholastic) and cognitive group (DLD, TD) as the between-subjects variables. Our initial analyses aimed to
6Our decision to allow participants in the scholastic training group to use either a tablet or a computer was based on the fact that both types of devices were available for use in the WM training group. It is merely by chance that the majority of the children in the scholastic training group used a computer for their sessions while the majority of the WM training participants used an iPad. While it may seem unfavorable for the two training groups to have primarily used different types of devices for their training, psychometric studies of device comparability show that device type does not significantly influence performance in children (Way, Davis, Keng, & Strain-Seymour, 2016).
9The Magic Memory data for four of the TD participants were lost for technical reasons and are not included in Figure 5.
7Only four of the five Magic Memory activities measured span capacity.
Activity 2 of the Magic Memory training program was ann-back task that trained participants’ability to update the contents of WM, and performance was measured in terms of percentage by the number of correct responses per block. Thus, the performance results of this task are not included in Figures 5 and 6.
8The Magic Memory data for six of the DLD participants were lost for technical reasons and are not included in Figure 4.
Figure 3.Illustration of a trial containing a sequence of three auditory stimuli from Activity 1, with the storage phase (the image on the left) and the reconstruction phase (the central image and the image on the right).
compare the pre- and posttest performances of the four groups separately. When composite scores were used, significant main effects of time were observed for both simple span,F(1, 71) = 65.79,p< .001,η2= .48, and complex span,F(1, 71) = 17.53,p< .001,η2= .20. Post hoc Tukey’s honestly significant difference (HSD) tests were used to investigate in which group(s) this effect was significant. It was shown that this effect was significant for both children with DLD (p< .01,d= 0.58 for simple span andp< .01, d= 0.49 for complex span) and TD children (p< .01, d= 0.45 for simple span andp< .01,d= 0.37 for complex span) in the WM training group, with posttest scores being significantly higher than pretest scores. Conversely, in the scholastic training group, there was no significant difference between pre- and posttest scores, neither for children with DLD nor for the TD children. These results are summarized in Table 6.
When WM tests were analyzed individually, a signifi- cant effect of time was observed for all six tests: serial order word span,F(1, 71) = 37.00,p< .01,η2= .34; forward digit recall,F(1, 71) = 29.00,p< .01,η2= .29; nonword repetition,F(1, 71) = 9.84,p< .01,η2= .12; backward digit recall,F(1, 71) = 5.27, p= .03,η2 = .07; running span, F(1, 71) = 7.97,p < .01,η2= .12; and counting span, F(1, 71) = 9.58,p < .01,η2= .12. Tukey’s HSD results showed that, for children with DLD in the WM training group, there was a statistically significant difference between pretest and posttest for serial order word span (p< .01,d= 0.50), forward digit recall (p< .01,d= 0.40), nonword repetition (p< .01,d= 0.46), and running span (p= .03,d= 0.55). For TD children in the WM training group, there was significant improvement in serial order word span (p< .01,d= 0.50), forward digit recall (p< .01, d= 0.45), and counting span (p= .02,d= 0.31). No significant
Figure 4.Illustration of a trial containing a sequence of three auditory stimuli from Activity 3, with the high-speed visual discrimination of quantity task immediately succeeding each stimulus. Participants heard an everyday sound (e.g., a ringing telephone) while looking at a neutral screen (an image of the main character from the application holding a hand to his ear). Immediately following the presentation of the auditory stimulus to be stored, the participants were asked to complete the unrelated visual processing task.
Figure 5.Magic Memory performance results from Session 1 to Session 24 for 20 of the 26 participants with developmental language disorder (DLD) on four of the five activities.
progress was observed for any of the WM tests for the children in the scholastic training group, neither for DLD nor for TD children. The results of the individual WM tests are summarized in Table 7.
Next, the repeated-measures ANOVAs were used to examine if the target training had a specific effect across the two cognitive groups (DLD, TD) on untrained WM tasks when compared to the scholastic training. There was a statistically significant interaction effect of time by type of training for both composite simple span,F(1, 71) = 10.74, p= .01,η2= .13, and composite complex span,F(1, 71) = 11.15,p= .01,η2= .14. Analyzing WM tests individually, three of six showed a significant interaction effect, with
the improvement being observed in the Magic Memory group: forward digit span,F(1, 71) = 13.17,p< .01,η2= .16;
backward digit span,F(1, 71) = 7.93,p< .01,η2= .10; and counting span,F(1, 71) = 5.70,p= .02,η2= .07. Levene’s test for equality of variances showed no significant dif- ference between any of the variables used to study the interactions.
The repeated-measures ANOVAs were also used to provide insight about whether there was a significant inter- action between training type and the presence of DLD. These analyses showed no interaction for composite simple span, F(1, 71) = 0.57,p= .45,η2= .01, or for composite complex span,F(1, 71) = 0.51,p= .48,η2= .01. However, when
Figure 6.Magic Memory performance results from Session 1 to Session 24 for 12 of the 16 typically developing (TD) participants on four of the five activities.
Table 6.Summary of pretest versus posttest comparisons for the composite scores for simple and complex spans.
Task Time
DLD, WM DLD, scholastic TD, WM TD, scholastic
M(%)
(SD) M(%)
(SD) M(%)
(SD) M(%)
(SD)
Simple composite span Pre 38.09
(9.82)
36.81 (9.20)
57.08 (13.66)
54.98 (10.15)
Post 46.11
(9.80)
39.02 (9.82)
65.45 (12.83)
59.72 (10.42)
Tukey’s HSD *** ns *** ns
Complex composite span Pre 23.98
(7.97)
20.81 (7.47)
42.24 (13.19)
41.09 (17.12)
Post 30.67
(11.01)
22.45 (7.94)
49.99 (16.43)
41.07 (15.60)
Tukey’s HSD ** ns ** ns
Note. DLD = developmental language disorder; WM = working memory; TD = typical development; HSD = honestly significant difference;
ns= not significant.
**p≤.01. ***p≤.001.