Deficits of Learning in Procedural Memory and Consolidation in Declarative Memory in Adults With Developmental Language Disorder

Abstract

Purpose

This study examined procedural and declarative learning and consolidation abilities in adults with developmental language disorder (DLD) relative to their typical language (TD) peers.

Method

A total of 100 young adults (age 18–24 years) with (n = 21) and without (n = 79) DLD participated across two sites. Performance measures on a recognition memory task and a serial reaction time task were used to assess declarative and procedural memory, respectively. Performance was measured shortly after learning (8 a.m.) and again after a 12-hr, overnight delay (8 a.m.).

Results

Linear mixed-effects modeling was used to examine the effects of time and group membership on task performance. For the serial reaction time task, there were significant effects of group (TD > DLD) and time (Day 1 > Day 2), but no interaction between them. For the recognition memory task, there was a significant interaction between group and time, driven by overnight gains in the TD group, combined with stable performance across days by those with DLD.

Conclusions

In procedural memory, adults with DLD demonstrate a learning deficit relative to adults without DLD, but appear to have comparable retention of learned information. In declarative memory, adults with DLD demonstrate a deficit in the overnight enhancement of memory retrieval, despite typical-like learning exhibited when tested shortly after encoding.

Supplemental Material

https://doi.org/10.23641/asha.13626485

Developmental language disorder (DLD) is a term that describes idiopathic disorder(s) of language (Bishop et al., 2017) and has an estimated prevalence of 7% of the U.S. population (Tomblin et al., 1997). DLD during childhood is often associated with the omission of grammatical morphemes and difficulties with syntactic constructions (see Leonard, 2014, for a review). An emerging literature suggests that, though individuals with DLD no longer appear to have obvious deficits in grammar by adulthood, they continue to struggle with various other language skills, such as narrative discourse, comprehension, and reading fluency (Del Tufo & Earle, 2020; Rost & McGregor, 2012; Suddarth et al., 2012). Moreover, there is mounting evidence that individuals with DLD continue to present with difficulties in language-related learning in adulthood (Earle et al., 2018; McGregor et al., 2017). Therefore, the age-appropriate use of grammar by adults with DLD may reflect compensatory processes rather than a recovery of core learning deficits that may in fact persist throughout the life span.

There is growing evidence that the language problems associated with DLD stem from general (nonlinguistic) deficits in learning and memory (fast mapping: Alt & Plante, 2006; procedural learning: Lum et al., 2014; sequence learning: Hsu & Bishop, 2014; statistical learning: Lammertink et al., 2017). The Procedural circuit Deficit Hypothesis (PDH; Ullman et al., 2020; Ullman & Pierpont, 2005) is a theoretical framework rooted in the neurobiology of DLD that offers a unifying account for deficits found across various types of learning. The PDH posits that the behavioral manifestations of deficits observed in individuals with DLD can be largely accounted for by abnormalities in brain structures that underlie procedural memory. According to this framework, learning and associated functions that rely on this circuitry are predicted to be impaired in DLD, while leaving declarative learning and memory functions relatively unaffected. The PDH is growing in empirical support within the literature on children with DLD (Hedenius et al., 2011; Lukács et al., 2017; Lum et al., 2012; see Lum et al., 2014, for a meta-analysis; cf. West et al., 2018; although, see Conway et al., 2019, for a reply to West et al., 2018). However, the procedural and declarative learning and memory abilities of adults with DLD are relatively unknown.

Beyond initial learning, there is growing evidence that consolidation of memory during sleep is also important for language learning (see Gómez et al., 2011, and Schreiner & Rasch, 2017, for reviews). To illustrate, sleep, but not a comparable period of wake state, has been observed to result in the integration of novel word forms into the mental lexicon (Gaskell & Dumay, 2003), as well as the generalization of speech sound categories (Earle & Myers, 2015) and grammatical rules (Gómez et al., 2006). These findings, taken together with nocturnal epileptiform activity observed in children with DLD, have prompted some to propose a causal relationship between sleep and language symptoms in the disorder (see Overvliet et al., 2010, for a review). However, the extant findings on memory consolidation in individuals in DLD are mixed, in which deficits have been found in some tasks (Adi-Japha et al., 2011; Earle et al., 2018; Hedenius et al., 2011) but not others (Lukács et al., 2017; McGregor et al., 2013, 2017).

A potential explanation for these inconsistent findings may be that studies have differed in the extent to which the consolidation effect under observation relies on sleep. Sleep is strongly associated with memory consolidation for declarative memory (Gais & Born, 2004). In contrast, the relationship between sleep and changes to behavior in procedural memory tasks is unclear (Nemeth et al., 2010; Walker et al., 2003). Therefore, in following the logic that consolidation deficits in DLD stem from hypothesized disruptions to sleep, it stands to reason that this phenomenon would be more likely to be observed in declarative rather than procedural memory performance.

The current study had two research aims. First, it aimed to examine procedural memory function assessed shortly after learning, and again after a 12-hr overnight delay, in adults with and without DLD. If, as hypothesized by the PDH, learning deficits in children with DLD are rooted in neuroanatomical abnormalities in procedural memory circuitry, the same learning deficits may be expected in adults with DLD. Furthermore, as sleep may not play a crucial role in memory consolidation in procedural memory, consolidation effects may be relatively conserved in those with DLD. Our second aim was to investigate declarative memory function assessed shortly after learning and again after a 12-hr overnight delay. We predicted that adults with DLD will demonstrate typical-like declarative memory when tested shortly after learning. However, we reasoned that the hypothesized interruptions in sleep-mediated processes would prevent overnight improvement in declarative memory retrieval in the DLD group (see Diekelmann & Born, 2010, for a review).

Method

Power Analysis

We used G*power software (Faul et al., 2007) to calculate our target sample size for the hypothesized within- and between-group interaction effect in declarative memory, assuming a Cohen's f of .59 and an alpha of .05. We assumed a large effect size for this analysis based on the large effect size observed for a similar analysis approach in Earle et al. (2018). This prompted us to recruit a sample of at least 20 participants per group. Additionally, we prospectively planned to follow up on our research aims with exploratory analyses, in order to determine the proportions of variance in language-related functions in adulthood that might be explained by differences in memory function. As these analyses do not directly address the stated objectives of this article, but are meant to further inform these findings, they are presented in Supplemental Material S1. To calculate the power for the exploratory analyses, we assumed an alpha of .05 and two predictors for a linear multiple regression analysis. We assumed medium effect sizes (Cohen's f values of .15–.2), based on previous relationships observed between memory consolidation and language ability (Hedenius et al., 2011). This calculation suggested that we needed between 81 and 107 participants. We thus aimed to recruit a total sample between 81 and 107 participants, with at least 20 participants in each group.

Participants

This study presents data collected across two sites, the University of Connecticut (UConn) and the University of Delaware (UD). Criteria for inclusion were that participants must be monolingual native speakers of English (aged 18–24 years) with typical (or corrected-to-typical) vision; no history of hearing impairment, cognitive impairment, or socio-emotional disorders; and no history of frank neurological insult or injury. In addition, individuals with documented histories of mood or attention disorders (including attention-deficit/hyperactivity disorder) were excluded from the study, on the basis that these conditions are known to alter sleep (Konofal et al., 2010; Tsuno et al., 2005). Finally, participants needed to meet criteria for inclusion in either the DLD or the typical language (TD) groups (see below).

Participants at UConn were recruited from the psychology department participant pool and were given course credit for participation in the study. Participants at UD were recruited through institutional review board–approved flyers and social media announcements, and were compensated $10/hr in gift cards for their time. At both sites, advertisement materials included language welcoming of those with a history of language learning disabilities, in order to recruit a greater representation of those with DLD than is found in the normal distribution. In order to control for the possible influence of the differences in recruitment procedures across sites on the resultant samples, we statistically control for data collection site in our analyses.

At UConn, 56 participants enrolled in the study met our inclusionary criteria. Of these, 15 participants did not complete all three sessions of the study. At UD, 88 participants enrolled in the study met our inclusionary criteria. Of these, 29 did not complete all three sessions (completion rate at UD may be low due to the fact that participants there were given an option to participate in this and a different study following the initial test session). Thus, we report data from 41 participants (32 TD, nine DLD) from UConn and 59 (47 TD, 12 DLD) from UD. The racial and ethnic distribution of our samples are as follows. In the UConn sample, 34 participants self-identified as White, three as Black, three as multiracial, and one declined to answer. Of these, one participant self-identified as Hispanic, 39 self-identified as non-Hispanic, and one declined to answer. In the UD sample, 50 participants self-identified as White, four as Black, three as Asian, and two as multiracial. Of these, five participants self-identified as Hispanic, and 54 participants identified as not Hispanic. All participants reported “some college” as their highest level of education. Please see Table 1 for sample characteristics, broken down by site. We note that the UConn sample is younger on average than the UD sample: This is likely because the psychology department participant pool at UConn included mostly first-year undergraduate students.

Table 1.

Characteristics of study samples.

		UConn		UD
		TD	DLD	TD	DLD
	N	32	9	47	12
	Age	18.94(.58)	19.12 (.76)	21.54 (2.63)	21.51 (2.15)
	Male:female	15:17	4:5	11:36	3:8
	Fidler et al. (2011, 2013) language measures
	Modified token test	40.25 (2.87)	36.89 (4.34)	39.55 (3.04)	33.75 (5.64)
	Spelling test	10.44 (1.71)	5.86 (1.25)	11.81 (1.78)	7.67 (2.90)
	Index	−0.93 (0.50)	0.50 (.36)	−1.14 (0.49)	0.52 (0.53)
	Word-level reading
Timed	Real word	96.88 (8.06)	98.11 (3.76)	94.89 (9.61)	87.17 (7.72)
	Pseudoword	57.47 (4.75)	52 (6.69)	59.62 (3.63)	52.33 (12.02)
Untimed	Real word	42.56 (1.94)	40.89 (2.02)	42.47 (1.67)	39 (2.73)
	Pseudoword	24.66 (1.21)	21.89 (1.62)	23.51 (1.55)	20.08 (3.29)

Note. Table 1 presents average raw scores and standard deviations by group (TD = typical language; DLD = developmental language disorder) and site (UConn = University of Connecticut; UD = University of Delaware). Word-level reading reflects performances on the Test of Word Reading Efficiency (UConn; Torgesen et al., 1999) and Test of Word Reading Efficiency–Second Edition (UD; Torgesen et al., 2012) subtests for timed, and the Woodcock Reading Mastery Test–Third Edition (Woodcock, 2011) subtests for untimed reading.

In order to be included in our DLD group, participants were required to self-report a history of receiving services and to be identified as language impaired according to the procedure for identifying adults with DLD devised by Fidler et al. (2011). To summarize this procedure, one enters the raw scores obtained from a modified token test and a spelling test (mainly irregular words, see Fidler et al., 2011, for the 15-item wordlist) into a discriminant function (index = 6.5727 + spelling × −.2184 + token × −.1298; Fidler et al., 2011, 2013), for which positive values indicate the presence of language impairment. The authors report sensitivity and specificity rates at 80% and 87% for this method, respectively, which seems to be emerging as the standard for the identification of adults with DLD (Bourgoyne & Alt, 2017; Earle et al., 2018; McGregor et al., 2017, 2020; Theodore et al., 2020; von Koss Torkildsen et al., 2013). To reduce the possibility of identifying poor spellers with good spoken language skills as DLD, we decided a priori to exclude individuals who had a positive value on this index, but who reported no prior history of spoken language problems and also met commonly accepted research criteria for dyslexia (1.5 SDs below the mean on at least two out of the four reading-level subtests administered at the time of the study; Gabrieli, 2009). In other words, individuals with dyslexia and no indication of DLD beyond a positive value on the index were excluded (n = 1). This, combined with the criteria above (e.g., no history of cognitive impairment or attention-deficit/hyperactivity disorder), resulted in a narrowly defined sample of 21 adults with DLD (in contrast to the broader, umbrella term for DLD described by Bishop et al., 2017).

In order to be included in our TD group, participants were required to report no history of services and to obtain a value on the Fidler et al. (2011) discriminant function that indicated no presence of DLD. As with our DLD sample, individuals who obtained reading scores that indicated the presence of developmental dyslexia were excluded. This resulted in a sample of 79 adults in our TD group.

Procedure

Overview

All participants provided informed consent prior to participation, according to procedures approved by the institutional review board at the respective sites. In order to confirm study and group eligibility, all participants first completed a 2-hr session consisting of a battery of cognitive and language tests administered by a trained experimenter in a small, quiet testing room. During this testing session, participants filled out a demographic and language history questionnaire. In addition, participants were scheduled for a two-session experiment that took place from 7:30 to 9 p.m. for Session 1 (Day 1) and 8 a.m. to 9 a.m. on the following day for Session 2 (Day 2). During the first experimental session, all participants completed the declarative memory task and the procedural memory task. During this session, participants were also given a perceptual memory task that is not discussed here; while uniform in design and task duration across the sites, it differed in the auditory stimuli that were used and, therefore, is not directly comparable across sites. Participants were assessed in their learning performance in the memory tasks at Day 1 and again on Day 2, after an approximate 12-hr overnight delay.

Test Administration

All test administration was conducted one-on-one with the experimenter. Participants were offered frequent breaks between the various subtests. Subtest scores were calculated from raw score sheets by two independent scorers for accuracy. Reliability between the two scorers for the present data set was above 97%. See Table 1 for performance on language and reading measures by group and site.

The following language assessments were administered. First, as mentioned above, DLD status was confirmed by the modified token test and the spelling test, described by Fidler et al. (2011, 2013). Measures to identify developmental dyslexia comprised four subtests of word-level reading ability. Untimed word-level reading was assessed using the Word Attack and Word Identification subtests of the Woodcock Reading Mastery Test–Third Edition (Woodcock, 2011). Timed word-level reading ability was assessed using the Sight Word Efficiency and Phonemic Decoding Efficiency subtests of the Test of Word Reading Efficiency (Torgesen et al., 1999) at UConn and the Test of Word Reading Efficiency–Second Edition (Torgesen et al., 2012) at UD. In addition to these, we administered additional assessments to better characterize our samples. These assessments include measures of nonverbal cognition (nonverbal IQ/perceptual reasoning index), reading comprehension, reading fluency, rapid automatized naming, verbal working memory, phonological processing, and executive function. Please see Supplemental Material S1 for a description of the tests administered and a report of mean scores obtained by group by site.

Experimental Procedures

Stimulus presentation and response recording were controlled via E-Prime 2.0 at the UConn, and E-Prime 3.0 at the UD (Psychology Software Tools, Inc.).

Experimental assessment of procedural memory. Procedural memory was assessed with the serial reaction time task (Nissen & Bullemer, 1987). This version of the task was adapted for E-Prime 2.0 (UConn) and E-Prime 3.0 (UD) from an earlier version used previously to assess procedural memory in children with DLD (Lum et al., 2012). The visual stimulus for this task was an image of a smiley face obtained from a clip art gallery. The serial reaction time task was chosen to index procedural memory function based on findings that procedural memory structures (e.g., basal ganglia) are critical for performance on this task (Clark et al., 2014; Janacsek et al., 2020).

Participants were presented with a dark screen, marked with the outline of four boxes arranged horizontally across the screen. They were asked to put four fingers on four marked, consecutive keys on a keyboard. They were told that a smiley face would appear in one of the four locations, and their task would be to hit the key corresponding to the location of the smiley face as quickly and as accurately as possible. During the Day 1 session, participants completed a warm-up block of 40 trials in which the smiley face was presented in a pseudorandom order. This was followed by four blocks of 80 trials in which the smiley face appeared in a repeated sequence of 10 locations. Participants then completed a final block of 80 trials in which the smiley face appeared in a pseudorandom order. During the Day 2 experimental session, participants again completed a warm-up block of 40 pseudorandom trials, followed by a single sequence block of 80 trials, and a final pseudorandom block of 80 trials.

Experimental assessment of declarative memory. Declarative memory was assessed using a recognition memory task. This task was adapted from an earlier version used to assess declarative memory in children with DLD (Lukács et al., 2017). The task was specifically developed to target learning by the declarative memory system (see Brown & Aggleton, 2001, for a review on the role of the perirhinal cortex and hippocampus on recognition memory), while reducing attentional and working memory demands on task performance (Hedenius et al., 2013; Lukács et al., 2017). In particular, this task that minimizes reliance on processes such as phonology, working memory, and recall that rely on circuits underlying procedural memory (Ullman et al., 2020; Ullman & Pierpont, 2005).

The visual stimuli in the task comprised 128 black-and-white drawings of objects. Of these, 64 were of real objects and 64 were of made-up objects (for details, see Earle et al., 2020). The real objects were matched for word frequency, number of syllables, and number of phonemes across the sets used for the encoding and test phases. The recognition memory task was administered in three phases: encoding, recognition, and retention. The encoding and recognition phases were administered during the Day 1 experimental session.

During the encoding phase, participants were instructed to place an index finger from each hand on marked keys on either side of the keyboard (“s” and “l”). They were told that they would see a series of objects, and they need to indicate, as quickly and as accurately as possible, if the objects were real or not real. Each trial began with a 1-s fixation cross at the center of the screen, followed by a 500-ms presentation of the target in the center, with written prompts (“real”/”made-up”) at the bottom corners of the screen. The left/right presentation of real/made-up was counterbalanced across participants. Trials ended at 500 ms if the response occurred during the stimulus presentation. If the response occurred after 500 ms, a fixation cross replaced the object until the participants indicated their response (max additional 4500 ms allowed per trial). This procedure ensured equal duration of exposure across items. Participants completed three practice trials, followed by 64 trials (32 real/32 made-up). Items were administered in a pseudorandom order in order to avoid more than three consecutive items of the same trial type (real/made-up).

The recognition phase was administered about 10 min after the encoding phase. Before the recognition phase, participants were told that they would see a series of objects and that some of the objects were seen before in the previous task, while others had not been seen. Participants were again asked to put an index finger from each hand on the marked “s” and “l” keys, and to respond as quickly and as accurately as possible to indicate whether the object was or was not seen before. The left/right presentation of seen/not seen was counterbalanced across participants. The trial structure was otherwise the same as in the encoding task, with written prompts “seen before?” beneath the target image and “yes”/”no” presented at the bottom corners of the screen. Following six practice trials, participants completed 128 trials of this task (64 seen/64 not seen before).

The retention phase was administered approximately 12 hr later, during the Day 2 experimental session. The items were administered in a different pseudorandom order from Day 1, and participants were given a different set of six practice trials and were presented with a new set of foils (i.e., items not seen before) in the experimental trials. Otherwise, the retention phase was identical to the recognition phase.

Data Overlap With Previous Publications

We note that the standardized test scores of our samples (scores presented in Table 1 and Supplemental Material S1, Table S1) are included in a larger data set that has appeared in previous publications on the skill profiles of college students with language-based learning disabilities (Del Tufo & Earle, 2020). The data on the experimental measures of declarative and procedural memory have not been presented in a peer-reviewed publication and is the original contribution of this article.

Analyses and Results

An anonymized version of the critical data set presented in this article, along with the R code used for the main analyses, is made publicly available via a GitHub repository (https://github.com/fsearle/EARLE-DMPM-Consolidation). These data are shared with the intent to promote transparency and replicability of research, and the authors request the proper attribution of the source for further educational or scholarly use of this material.

Data Transformations

In order to index individual performance on the serial reaction time task, we first removed outlier (i.e., reaction times greater than 3 SDs from the by-participant, by-block mean) trials and trials in which the response was incorrect. We then calculated the average reaction times in the last sequence block and the subsequent random block, both on Day 1 and on Day 2. We then calculated the difference between the reaction times for the random and sequence blocks, to arrive at a procedural memory score for each Day per participant.

In order to index individual performance on the recognition memory task, we first calculated the d' score (MacMillan & Creelman, 2004) from the responses in the recognition and retention phases of the task. A d' score was first calculated for the real and made-up objects separately, in case response bias differed between real and made-up items. These scores were then averaged across item types, in order to arrive at a single declarative memory score for each day per participant. Please see Table 2 for descriptive statistics of task performance by group.

Table 2.

Experimental task performance by group.

		TD	DLD
		n = 79	n = 21
Serial reaction time performance
Day 1	Random	483.00 (75.85)	537.27 (108.20)
	Sequence	434.30 (83.60)	512.33 (163.34)
	Difference	48.70 (39.99)	24.94 (74.41)
Day 2	Random	426.29 (58.63)	480.92 (106.01)
	Sequence	390.49 (62.20)	462.00 (109.11)
	Difference	35.80 (26.92)	18.97 (44.38)
Recognition memory performance
Day 1	Real	1.88 (1.05)	1.76 (1.29)
	Made up	1.07 (1.09)	1.02 (1.27)
	Average	1.47 (0.93)	1.39 (1.11)
Day 2	Real	2.27 (1.05)	1.84 (1.33)
	Made up	1.41 (1.11)	1.23 (1.24)
	Average	1.84 (0.96)	1.48 (1.25)

Note. Table 2 describes the average and standard deviations of task performance. For the serial reaction time task, the values expressed are the average reaction times (in milliseconds) of all accurate trials in the last random block (Random) and the last sequence block (Sequence) of each day. For recognition memory performance, the values are expressed in d' (MacMillan & Creelman, 2004) for recognition trials on real objects (Real) and made-up (Made Up) objects. TD = typical language; DLD = developmental language disorder.

In order to ensure that our measures of procedural and declarative memory are on commensurate scales for the individual differences analyses (see Supplemental Material S1), the outcome variables for the serial reaction time task and the object recognition memory tasks were transformed according to the proximity-to-maximum scaling method (Moeller, 2015). This method is preferred over Z-standardization for repeated-measures designs, so as to conserve both the between-subjects variability within each time point, and also the within-subject variability across time.

Due to equipment malfunction, we experienced data loss from two participants on Day 1 procedural memory and from one participant on Day 2 declarative memory. Missing cases were estimated during analysis using the Multivariate Imputation by Chained Equations (package “mice”; van Buuren et al., 2015) in R (R Development Core Team; Version 3.4.1, 2017).

Group and Posttraining Delay Effects on Procedural Memory

We first tested the skewness and kurtosis of our data with the D'Agostino and Bonett tests using the “moments” package in R (Komsta & Novomestky, 2015). We found that the p.m. scores were negatively skewed on both days, Day 1: skew = −.97, z = −3.64, p < .001, tau = .14, z = 1.53, p = .124; Day 2: skew = −.97, z = −3.65, p < .001, tau = .11, z = 1.56, p = .120. Therefore, we employed a generalized linear mixed-effects modeling approach, specifying a gamma family of distributions for our outcome measure. The generalized linear mixed-modeling approach to repeated-measures analysis is more robust to unequal sample sizes than an analysis of variance (VanLeeuwen et al., 1996). In order to determine the effect of group (TD/DLD), time (Day 1/Day 2), and the interaction between group and time on procedural memory performance, a mixed-effects model was fitted on the procedural memory scores as the outcome measure, with the three fixed factors mentioned above and participant and study site entered as random factors. One to zero coding was used for our fixed factors, referenced to UConn for school, Day 1 for time, and to TD for group. Models were fitted using the lme4 package (Bates et al., 2015) in R (R Development Core Team, 2017). Below, we report standardized coefficients for the predictors in our overall model, Akaike information criterion (AIC) = −34.2, Bayesian information criterion (BIC) = −11.1, LogLikelihood = 24.1, r ² _m = .102, r ² _c = .501. Group and time were significant factors, group: β = −.13, SE = .05, p = .010, partial Rsq = .061; time: β = −.05, SE = .02, p = .011, partial Rsq = .019, but the interaction between time and group was not significant, β = .04, SE = .02, p = .132, partial Rsq = .006. These effects suggest that, across both groups, the participants obtained lower procedural memory scores on the serial reaction time task after a 12-hr delay. In addition, the participants with DLD performed worse than the TD group across both days. See Figure 1 for a graphical depiction of these results.

Figure 1.

Serial reaction time task performance by day by group. Box and whisker plots of serial reaction time (SRT) performance (average reaction time of random trials minus sequence trials) by day by group. Boxes depict second and third quartiles, with the median represented by the horizontal line. Whiskers depict first and fourth quartiles, with outliers represented by single points. DLD = developmental language disorder; TD = typical language.

Group and Posttraining Delay Effects on Declarative Memory

Our initial tests of skewness and kurtosis of our data were not statistically significant, Day 1: skew = −.21, z = −.89, p = .372, tau = .080, z = 1.23, p = .22; Day 2: skew = −.22, z = −.94, p = .347, tau = .09, z = .47, p = .641. In order to determine the effects of group, time (overnight delay), and their interaction on declarative memory performance, we fitted a linear mixed-effects model (that assumes a Gaussian distribution) for the recognition memory task d' scores as the outcome measure, with the three fixed factors and participant and study site entered as random factors. Again, we report standardized coefficient for the resultant model, AIC = −331.1, BIC = −308.1, LogLikelihood = 172.6, r ² _m = .067, r ² _c = .502. The model included a significant effect of time, a significant interaction between group and time, but no effect of group, time: β = .025, SE = .005, p < .001, partial Rsq = .050; Group × Time: β = −.014, SE = .007, p = .041, partial Rsq = .012; group: β = −.003, SE = .009, p = .735, partial Rsq = .001. Follow-up paired-samples t tests revealed the source of the interaction as a significant increase in recognition memory performance between Day 1 and Day 2 for the TD group, t(78) = −4.65, p < .001, 95% CI [−0.079, −0.032], Cohen's d = .57, but no significant performance differences between the 2 days in the DLD group, t(19) = 0.10, p = .925, 95% CI [−0.063, 0.069], Cohen's d = .03. This indicates that while there were no overall weaknesses in declarative memory associated with the DLD group, the pattern of changes on recognition memory differed between the groups. Specifically, while those in the TD group made significant gains in recognition memory overnight, individuals with DLD did not. See Figure 2 for graphical summary of this pattern.

Figure 2.

Recognition memory task performance by day by group. Boxplots of recognition memory (RM) performance by day by group. Performance is expressed in d' (MacMillan & Creelman, 2004). DLD = developmental language disorder; TD = typical language.

The above findings suggest group-level differences in task indexes of procedural and declarative learning and retention between adults with and without DLD. The extent to which procedural and declarative memory are related to language-related abilities in adulthood in general is unclear. Thus, in exploratory analyses, we additionally examined the relations between individual differences in measures of memory and performance on language-related scores; see Supplemental Material S1.

Discussion

The current study investigated performance on procedural and declarative memory tasks in adults with and without DLD. Memory was assessed immediately after learning and again after a 12-hr overnight delay. Our main finding for procedural memory was a main effect of group, driven by better performance by the TD group than the DLD group across both days (see Figure 1). Our main finding for declarative memory was an interaction between group and time on performance, driven by performance gains on Day 2 by the TD group that were absent in the DLD group (see Figure 2). Additionally, exploratory analyses (see Supplemental Material S1) revealed that individual differences in performance on the memory tasks were associated with performance on the modified token task and the spelling test.

The finding that adults with DLD were impaired on a procedural memory task is consistent with the previous literature on children with DLD (see Lum et al., 2014, and Clark & Lum, 2017, for meta-analyses). This finding indicates that the procedural memory deficit often observed in children with DLD does not appear to be resolved in young adults with DLD. Together with the tendency for the hallmark linguistic symptoms of DLD (e.g., impoverished grammar; Leonard, 2014) to not persist into adulthood, the present finding suggests two things. First, it suggests that the disappearance of these hallmark linguistic symptoms may be masking a core impairment in learning that persists into at least young adulthood in individuals with DLD. Second, it suggests that language intervention, and/or intrinsic compensation by individuals with DLD, are in fact often successful in eventually establishing the target linguistic structures.

Procedural memory scores in the serial reaction time task did not improve following an overnight delay in either group. This observation is not surprising, since off-line gains in the knowledge of learned motor sequences have been only inconsistently observed in the prior literature (Kirov et al., 2015; Nemeth et al., 2010). Note also that we observed faster overall reaction times (across both sequence and random conditions) in both groups on Day 2 (see Table 2). This pattern, that is, a lack of gains in sequence-specific knowledge, combined with an overall enhancement in task speed, has been previously observed to take place over time, whether or not this time includes a period of sleep (Nemeth et al., 2010). Furthermore, the DLD–TD gap in sequence knowledge remained stable overnight, suggesting that the two groups demonstrated comparable retention of procedural information. Taken together, this pattern may suggest that individuals with DLD have comparable retention of procedural memory relative to TD adults. We discuss this finding in the context of the previous literature below.

A different pattern was observed for our declarative memory findings. First, the TD group made overnight gains in recognition memory performance in the absence of further training. This pattern is consistent with sleep-mediated consolidation effects that are often observed for declarative memory tasks (see Gais & Born, 2004, for a review). Second, these overnight gains were not observed in the DLD group, despite comparable performance between the TD and DLD groups in declarative memory when assessed shortly after learning. This finding suggests that the consolidation of declarative memory is impaired in adults with DLD.

At face value, the present finding that declarative memory consolidation is impaired in adults with DLD suggests a different interpretation from that of prior observations on novel word learning by McGregor et al. (2013, 2017). In McGregor et al. (2013, 2017), adults with DLD and TD were tested on word form recall (McGregor et al., 2013), word meaning recognition (McGregor et al., 2013), and word form stem completion (McGregor et al., 2017) following novel word learning. Over the course of the subsequent week, ¹ the performance deficit of the DLD group as compared to the TD group widened for word form recall (McGregor et al., 2013), but remained stable for word meaning recognition (McGregor et al., 2013) and word form stem completion (McGregor et al., 2017). Collectively, these results were interpreted as a weakness in encoding lexical forms, combined with intact retention of information underlying performance on word meaning recognition and stem completion, in adults with DLD.

In light of this discussion, we find our results to contribute to an emerging pattern of observations across studies. First, our findings indicate that adults with DLD retain learned information both in procedural and declarative memory, consistent with McGregor et al. (2013, 2017). Moreover, the present observation that overnight gains in recognition memory performance is diminished in adults with DLD echoes the pattern of performance on the word form recall task in McGregor et al. (2013) demonstrated by adults with DLD, as well as previous reports on learning in other domains by both adults and children with DLD (Adi-Japha et al., 2011; Earle et al., 2018; Hedenius et al., 2011). Taken together, individuals with DLD may have a deficit in the off-line improvement function of memory consolidation.

Memory consolidation is a term that covers various effects of time on memory following learning, with distinct underlying mechanisms for different types of changes to the memory trace (see Dudai, 2012, for a review). Off-line improvement of memory retrieval is a behavioral phenomenon attributed to “systems” consolidation, which has been posited to occur optimally during slow-wave sleep (see Diekelmann & Born, 2010, for a review). In contrast, consolidation processes responsible for localized changes to neural structures (“synaptic” consolidation, which assists in the retention of learned information as well as in enhancing task speed) occur regardless of whether or not that interval contains sleep (see Klinzing et al., 2019, for a review). Thus, the observation that while individuals with DLD appear to retain learned information, and yet do not demonstrate an off-line enhancement of recognition memory performance, may suggest that individuals with DLD are impaired in consolidation processes that occur during sleep. This interpretation is consistent with prior observations of atypical sleep in children with DLD (see Overvliet et al., 2010, for a review).

While disruptions in sleep-mediated consolidation is not a prediction that has been previously specified under the PDH (Ullman et al., 2020; Ullman & Pierpont, 2005), we note that these findings are compatible with this framework. Disruptions in sleep, and sleep-mediated processes, may be in fact a consequence of abnormalities in procedural memory structures. Specifically, the nucleus accumbens, a key structure for reward signaling during feedback-based learning, is posited to play an intrinsic role in the sleep–wake regulatory network (Lazarus et al., 2013). Consequently, neurodegenerative diseases of the basal ganglia have known associations with poor quality sleep (e.g., Huntington's disease [Morton, 2013], Parkinson's disease [Garcia-Borreguero et al., 2003]), marked by decreases in slow wave sleep and increased sleep disturbances. Indeed, both the nucleus accumbens (Lee et al., 2013) and sleep quality have been found to be atypical in children with DLD (see Overvliet et al., 2010, for a review). Thus, the challenges to learning experienced by those with DLD may be amplified by potential disruptions to sleep-mediated functions that support learning and memory.

Another prior finding that is important to address was reported by Lukács et al. (2017), who used an adapted version of the same recognition memory task as was used in this article. Lukács et al. (2017) found that Hungarian children with DLD made larger overnight gains on recognition memory than their typical peers, leading the authors to conclude that children with DLD may show enhanced declarative memory consolidation. It is unclear at this juncture how this and this study arrived at such different results using very similar tasks. One possibility is that consolidation patterns in declarative memory differ between children and adults with DLD. Sleep spindles, which are linked to declarative memory consolidation during sleep, are observed to mature during early adolescence (Hahn et al., 2019). Thus, patterns in sleep-mediated consolidation across those with and without DLD may differ to a greater extent in adulthood. We note that the Lukács et al. (2017) study also differed from this study in that, in the latter, the time of day of learning, of testing, and of retesting were controlled across participants. In addition, there were some differences in sample selection, including the exclusion of those with comorbid developmental dyslexia in the present sample. This raises an intriguing possibility that individuals with different comorbidities/DLD subtypes may differ in their pattern of off-line memory consolidation. This is an important possibility to pursue in the future.

There are several important limitations to acknowledge in this article. First, data collection occurred across two sites, under two slightly different recruitment procedures. The potential problems stemming from this were mitigated by the comparable numbers of participants in the TD and DLD groups from each site, together with our inclusion of site as a random effect in our analyses. In addition, the criterion that participants must be native monolingual speakers of English likely skewed the racial and ethnic distribution of our study sample. A similar limitation is that our basis for inclusion in the DLD group required participants to have accessed services in the past: This criterion may have introduced socioeconomic bias into our DLD sample. An additional limitation in the current study is that we lacked a comparison group that remained awake for the between-session interval. We opted instead to administer the learning tasks in the evening to all participants, in order to control for potential diurnal effects on learning. The interval scheduled between 8 p.m. and 8 a.m. further helped to minimize the differences in intervening wake-state activity levels of participants. Thus, while the overnight improvement in declarative memory performance is consistent with a sleep-mediated effect from the wider memory consolidation literature, we lack the data in this article to make that case definitively. Finally, the current data set lacks the measures to compare sleep quality or habits between groups. These will be important issues to address in future studies.

Despite these limitations, we suggest that the current findings have important implications. First, the study suggests that adults with DLD continue to present with learning challenges in adulthood, which may interfere with their academic attainment and employment outcomes. Second, even if individuals with DLD do not present with initial learning impairments, this does not guarantee that the subsequent consolidation of that memory trace will also occur typically. Specifically, individuals with TD abilities appear to make gains in memory on the next day without additional exposure, whereas those with DLD do not appear to do so. This may suggest that those with DLD must be exposed more often to the target information than those with TD abilities in order to establish comparable long-term representations. Finally, this work joins a growing call for the need to attend to sleep in clinical practice in speech-language pathology (McGregor & Alper, 2015; Morrow & Duff, 2020).

In summary, the current study identified deficits in procedural learning and in the off-line consolidation of declarative memory in young adults with DLD. These findings suggest a further theoretical refinement of the PDH framework, in that an important consequence of the hypothesized anomalies to procedural memory circuitry may be disruptions to sleep and to sleep-dependent memory consolidation. The study also has implications for clinical practice, in that individuals with DLD appear to present with deficits in learning and memory even after the successful remediation of obvious linguistic symptoms. Future work will examine relationships between sleep quality and off-line memory consolidation in young adults with DLD, as well as related potential strategies for intervention.

Funding Statement

The portion of this work that was carried out at the University of Connecticut was supported by the National Institutes of Health F31DC014194 to F. S. E. The portion of this work that was carried out at the University of Delaware was supported by National Institutes of Health R21DC016391 to F. S. E. and faculty start-up funding from the University of Delaware.