Verbal Working Memory Error Patterns and Speech-Language Outcomes in Youth With Cochlear Implants

Abstract

Purpose:

Verbal working memory (VWM) delays are commonly found in prelingually deaf youth with cochlear implants (CIs), albeit with considerable interindividual variability. However, little is known about the neurocognitive information-processing mechanisms underlying these delays and how these mechanisms relate to spoken language outcomes. The goal of this study was to use error analysis of the letter–number sequencing (LNS) task to test the hypothesis that VWM delays in CI users are due, in part, to fragile, underspecified phonological representations in short-term memory.

Method:

Fifty-one CI users aged 7–22 years and 53 normal hearing (NH) peers completed a battery of speech, language, and neurocognitive tests. LNS raw scores and error profiles were compared between samples, and a hierarchical regression model was used to test for associations with measures of speech, language, and hearing.

Results:

Youth with CIs scored lower on the LNS test than NH peers and committed a significantly higher number of errors involving phonological confusions (recalling an incorrect letter/digit in place of a phonologically similar one). More phonological errors were associated with poorer performance on measures of nonword repetition and following spoken directions but not with hearing quality.

Conclusions:

Study findings support the hypothesis that poorer VWM in deaf children with CIs is due, in part, to fragile, underspecified phonological representations in short-term/working memory, which underlie spoken language delays. Programs aimed at strengthening phonological representations may improve VWM and spoken language outcomes in CI users.

Early cochlear implantation in children with profound deafness—made possible by universal newborn hearing screening and expanded indications for cochlear implantation—provides auditory sensory input during critical periods of speech and language development. Following cochlear implantation, average rates of expressive and receptive language acquisition are greater than would be predicted by baseline (preimplant) measures, especially when implantation is performed at very early ages (i.e., < 18 months of age; Geers & Nicholas, 2013; Niparko et al., 2010). Indeed, many children who receive cochlear implants (CIs) at very young ages achieve near-normal speech perception in quiet, ideal listening conditions (Davidson et al., 2011). However, on average, children with CIs lag behind their normal hearing (NH) peers on measures of speech perception in noise (Davidson et al., 2011) and higher order language processing, such as morphological and syntactic structure, listening comprehension, and connected speech (Geers et al., 2009; Geers & Nicholas, 2013; Geers et al., 2003). Moreover, speech-language outcomes have been found to be highly variable among CI users, with some children performing at or above age-based norms and others performing well below average (Castellanos et al., 2016; Davidson et al., 2011; Geers et al., 2009; Geers & Nicholas, 2013; Geers et al., 2003; Niparko et al., 2010).

Better spoken language outcomes in children with CIs are significantly associated with conventional audiological and hearing history variables, such as older age at onset of deafness (Geers et al., 2003), better residual hearing in the form of lower preimplant pure-tone average (PTA; Geers & Nicholas, 2013), and younger age of implantation (Geers et al., 2009; Geers & Nicholas, 2013). Higher scores on measures of speech and language in children with CIs have also been associated with higher nonverbal intellectual ability (Geers et al., 2009; Geers & Nicholas, 2013; Geers et al., 2003) and demographic characteristics, such as female gender (Geers et al., 2009; Geers et al., 2003), higher household income (Kronenberger et al., 2014), higher parental education (Geers et al., 2009), and smaller family size (Geers et al., 2003). However, the combined effects of hearing history and demographic variables account for, at most, only about 50% of the observed variability in spoken language outcomes among pediatric CI users (Geers et al., 2009; Geers & Nicholas, 2013; Geers et al., 2003). The factors that contribute to the remaining variability are of great importance as potential targets for novel intervention programs aimed at preimplantation optimization or postimplantation rehabilitation, with the ultimate goal of improving speech and language outcomes after cochlear implantation.

Verbal Working Memory and Spoken Language Development After Cochlear Implantation

Because language development supports, and is supported by, other domains of neurocognitive ability (e.g., concept formation, reasoning, memory, and controlled attention), individual differences in neurocognitive functioning may underlie variability in language outcomes in CI users (Kronenberger & Pisoni, 2020). In particular, verbal working memory (VWM), the ability to concurrently hold verbal contents in short-term memory (STM) during other cognitive operations, has shown the most evidence of reciprocal associations with speech and language outcomes in deaf children with CIs (Kronenberger & Pisoni, 2018). Measures of VWM are significantly correlated with measures of spoken language in CI users, including reading (Fagan et al., 2007), vocabulary (Fagan et al., 2007), listening comprehension (Kronenberger et al., 2014), and speech perception (Cleary et al., 2001; Kronenberger et al., 2014).

Longitudinally, VWM predicts later language skills in preschool-age children with CIs, and this predictive association is stronger for CI users than NH peers (Kronenberger et al., 2020). Measures of VWM in elementary school are also associated with speech and language outcomes in middle and high school (Pisoni et al., 2011). Preliminary support for a causal influence of VWM on language outcomes has been provided by studies that demonstrate improvement in speech and language outcomes for pediatric CI users after working memory training (Ingvalson et al., 2014; Kronenberger et al., 2011) and by studies that show weaker performance on language tasks by CI users when working memory is taxed by a distracting task (Kronenberger et al., 2018). Conversely, measures of expressive vocabulary in the early postimplantation period are predictive of long-term VWM outcomes (Castellanos et al., 2016), while speech intelligibility in preschool is predictive of later working memory outcomes (Castellanos et al., 2014). These latter findings suggest that language and VWM are reciprocally associated in CI users, with each facilitating the growth of the other throughout child and adolescent development.

Empirical evidence and theoretical models suggest several potential reasons for the reciprocal associations between language and VWM in CI users: CI users show a much wider range of language and VWM outcomes than their NH peers, ranging from near-normal to significantly delayed (Niparko et al., 2010), and this wide range of outcomes allows for greater reciprocal influences of language and VWM at extreme values. Evidence supporting this hypothesis has been found in the form of stronger associations between language and VWM in CI users compared with NH peers (Kronenberger et al., 2014). Furthermore, VWM can be used to compensate for delays in fast, automatic language processing in deaf children with CIs by facilitating slower, more effortful processing that requires holding and processing linguistic information in short-term immediate memory (Rönnberg et al., 2013). In addition to VWM supporting language skills, language serves as a useful tool for encoding and maintaining information in VWM, resulting in enhanced VWM in children with stronger language skills (Alderson-Day & Fernyhough, 2015). As a result, VWM has been identified as both a foundational domain of neurocognitive risk and as a potential target for intervention to enhance speech-language outcomes in CI users.

Factors Explaining VWM Delays in CI Users

Because development of executive functions such as VWM begins by age 2 years (Cuevas & Bell, 2014), early auditory and language deprivation experienced by children with prelingual deafness may adversely affect an early and sensitive period of VWM development (Thompson & Steinbeis, 2000). Consistent with this hypothesis, prior research consistently demonstrates VWM delays, on average, in samples of CI users beginning at young ages. Pediatric CI users underperform NH peers on the digit span forward and digit span backward tests (Burkholder & Pisoni, 2003; Fagan et al., 2007; Nittrouer et al., 2013), two well-established measures of STM and VWM capacity. Studies using other measures of VWM have also found delays in both storage and processing of VWM in children with CIs (Nittrouer et al., 2017), and these delays have been replicated in other cultures and languages (Soleymani et al., 2014). VWM delays in children with CIs are found even when delays in visual working memory are not present (Bharadwaj et al., 2015). Findings of VWM delays in CI users cannot be entirely explained by deficits in perception and audibility of spoken stimuli or in the quality of speech production, as significant delays remain even when digits are presented visually and when responses are entered manually (AuBuchon et al., 2015). As with speech and language outcomes, there is substantial variability in VWM outcomes among children with CIs (Cleary et al., 2001; Pisoni et al., 2011), which is only partially explained by demographic and developmental hearing history experiences such as duration of deafness and preimplantation hearing quality (Castellanos et al., 2014; Kronenberger et al., 2014).

Although many studies have documented delays in VWM and strong associations between VWM and spoken language in CI users, very little is known about the elementary neurocognitive information-processing operations that underlie these delays. One explanation is that slow verbal rehearsal speed (the speed of subvocal rehearsal of words in active working memory) in CI users reduces the number of stimulus items that can remain activated in VWM, resulting in loss of memory for items that cannot be rehearsed quickly enough (Pisoni et al., 2011). Rehearsal speed has been well documented as a significant factor in VWM capacity in NH populations, and manipulations that reduce rehearsal speed result in poorer working memory performance (Cowan & Kail, 1996). An additional speed-based factor specifically at risk in children with CIs is scanning speed of STM, as indexed by much longer durations (time length) between digits spoken during digit span recall (“pause duration”) in CI users compared with NH peers, suggesting slower retrieval speed from verbal STM (Burkholder & Pisoni, 2003). The role of rehearsal speed and STM scanning speed for explaining VWM delays in children with CIs has been supported by studies that demonstrate associations between verbal rehearsal speed, pause durations, and digit span scores in CI users (Burkholder & Pisoni, 2003; Pisoni et al., 2011).

Because verbal rehearsal speed and STM scanning speed ultimately reflect the fluency/robustness of phonological coding (Pisoni et al., 2011), the adverse effects of slow verbal rehearsal speed and long pause durations on VWM delays in CI users may be a result of fragile, underspecified representations of words in STM. Prior work in NH children demonstrates that the quality of phonological storage in STM underlies the ability to retrieve fine phonological details in nonword repetition tasks (Gathercole, 2006). Similarly, in children with CIs and NH, phonological processing quality was a better predictor of nonword repetition performance than STM of phonologically distinct word lists (Nittrouer et al., 2014). Thus, deficits in rapid phonological coding and phonological representations may underlie VWM delays in CI users.

Internal phonological representations in memory that are coarsely coded and less refined at the time of encoding and registration (i.e., the memory codes are fragile/underspecified) are more susceptible to memory failure from decay (forgetting) or interference (substituting a like-sounding word). For example, in both CI and NH participants, verbal memory performance is worse for lists of words that are phonologically similar than for lists of words that are phonetically dissimilar (the “phonological similarity effect”; Conrad, 1964; Cowan & Kail, 1996; Nittrouer et al., 2013). Additionally, studies in NH participants have shown that incorrectly recalled letters on visually presented letter span tests are not random but, rather, probabilistically determined by their level of acoustic confusability with the target item (Conrad, 1963). Because these confusions occur in visually presented letter spans, stimulus audibility cannot explain this effect. The adverse effect of fragile/underspecified phonological representations on VWM is likely to be stronger in pediatric CI users, who have experienced an extended period of little or no auditory input, followed by exposure to the coarsely encoded auditory input of a CI. However, apart from a single study by Nittrouer et al. (2013), there has been little investigation into phonological similarity effects on VWM in CI users. No research has compared phonological similarity effects with other factors that may influence performance (e.g., sequencing errors, additions, and omissions) using a challenging working memory task requiring multiple mental operations concurrent with verbal STM in samples of CI users compared with NH peers.

Advantages of the Letter–Number Sequencing Test for Explaining VWM Delays in CI Users

The letter–number sequencing (LNS) subtest of the Wechsler Intelligence Scale for Children–Fifth Edition (WISC-V; Wechsler, 2014) is a widely used test of VWM, in which subjects are presented with a random, mixed series of letters (A–Z) and digits (1–9; e.g., R-5-J-3-X) and instructed to recall the digits in ascending order, followed by the letters in alphabetical order (3-5-R-J-X). This task requires encoding and retention of two types of verbal stimuli (digits and letters) in verbal STM, concurrent with the mental operations of separating and independently reordering stimuli on output according to two different rules (ascending order for digits and alphabetical order for letters). Thus, the LNS test places a considerably greater cognitive load on individuals than conventional digit span or word recall tests and appears to draw on the neurocognitive processes of attention, visuospatial function, and processing speed (Crowe, 2000). Moreover, the LNS test is uniquely suited for investigating the information-processing mechanisms underlying VWM performance. Because the LNS test uses phonologically confusable stimuli (Cowan & Kail, 1996) and requires multiple processing operations, several types of well-defined response errors can occur, allowing for identification of several possible causes of working memory errors: order errors (digits and/or letters out of order), phonological errors (substitution of an incorrect digit/letter for a phonologically similar target digit/letter), commission errors (substitution of a digit/letter for a digit/letter with which it does not share phonological similarity), omission errors (omission of a target digit/letter from the recalled sequence), and perseveration errors (successive repetition of a correctly recalled item; Mielicki et al., 2018).

Despite the value of LNS error analysis for understanding the cause of VWM errors, LNS error analysis has not been applied to prelingually deaf youth (defined for purposes of this study as school-age [7–12 years], adolescent [13–17 years], and late adolescent/early adult [18–22 years] age ranges, corresponding to ages during which scores on VWM tests show consistent improvement; Schrank et al., 2014) with CIs. Error analysis of VWM and STM tests in youth with CIs has been limited to digit span tests (Burkholder & Pisoni, 2004). Because digits are minimally phonologically confusable, error analyses of digit span tests are unable to provide a strong test of hypotheses about the potential effects of fragile, underspecified phonological representations of stimuli on VWM performance. Furthermore, existing LNS error coding procedures provide insufficient systematic and specific rules and applications for reliably encoding and interpreting error patterns in order to explain VWM delays in CI users (Mielicki et al., 2018). Thus, in order to better understand and explain VWM development in CI users, the goals of this study were as follows: (a) to propose a systematic, reliable procedure for coding errors on the LNS subtest; (b) to compare the LNS performance and error profiles of youth with CIs with those of NH peers; and (c) to determine the association between LNS accuracy and error patterns with measures of hearing, speech, and language outcomes.

This study tested three hypotheses based on previous findings of VWM in CI users (Kronenberger & Pisoni, 2020): (a) The CI sample will achieve a lower average LNS raw accuracy score, compared with the NH sample, reflecting poorer VWM skills. (b) The CI sample will show a higher rate of phonological errors than the NH sample, reflecting fragile phonological representations of words in STM. (c) Measures of functional hearing, rapid phonological coding in speech perception, and language will correlate positively with LNS raw accuracy scores and negatively with phonological error scores, reflecting the central role of VWM and robust encoding of phonological representations in hearing, speech, and language outcomes. Identifying the mechanisms of action underlying the well-documented VWM delays in CI users is a crucial step toward explaining variability and individual differences in neurocognitive processes on spoken language outcomes after implantation and developing novel intervention programs to address these at-risk outcomes.

Method

Participants

Participants in this study were 51 youth with CIs and 53 NH peers. Inclusion criteria for CI users were as follows: (a) moderate-to-profound, bilateral hearing loss diagnosed at < 3 years of age; (b) cochlear implantation at ≤ 3 years 11 months of age; (c) CI use for > 7 years; and (d) an auditory–oral communication mode, defined by Geers and Brenner (2003) as “lipreading and listening both encouraged, sign language rarely used.” Inclusion criteria for NH participants were as follows: (a) normal hearing, speech, and language by parent report and (b) a normal hearing screen (where the right and left ears were tested separately with 500-, 1000-, 2000-, and 4000-Hz tones presented at 20 dB via Telephonics TDH-50P headphones in an Acoustic Systems RE-243 sound booth). Inclusion criteria for both samples were as follows: (a) no developmental, cognitive, or neurological diagnoses; (b) a home environment in which spoken English was the primary language; (c) age 7–22 years; and (d) nonverbal IQ ≥ a scaled score of 4 (i.e., 2 SDs below the normative mean) on the Classification & Analogies subtest of the Leiter International Performance Scale–Third Edition (Leiter-3 CA; Roid et al., 2013).

Youth between ages 7 and 22 years were included in order to obtain a large sample across the developmental period between school age and late adolescence/early adulthood (corresponding to a period of growing VWM skills; Schrank et al., 2014). This broad age range also allowed for the investigation of differences in LNS error scores across age groups as well as of Age Group × Hearing Group (CI vs. NH) interactions. It is well established that the number and difficulty of total LNS sequences recalled improve with age (Wechsler, 2014), and VWM skills improve throughout the 7- to 22-year age range (Schrank et al., 2014). However, there have been no prior investigations of age differences or Age Group × Hearing Group interactions with types of LNS errors. Three age groupings were created to investigate these potential age effects: school-age (7–12 years), adolescent (13–17 years), and late adolescent/early adult (corresponding approximately to college ages: 18–22 years).

Procedure

Study procedure was approved by the local institutional review board, and written informed consent was obtained from parents (with the assent of children as appropriate) or adult participants. All tests were administered during one session in a standardized order without additional modifications/accommodations and per standard instructions by speech-language pathologists who were certified by the American Speech-Language-Hearing Association (ASHA) and experienced in the evaluation of children and adolescents with CIs. Spoken items were administered with participants seated in full view of the examiner's face, whereas audio-recorded test items were presented at 65 dB SPL using a high-quality loudspeaker located approximately 3 ft from participants. Tests were recorded with audio and/or video to check scoring later, if necessary. For participants who attended the session with parents, the Quality of Hearing Scale (QHS; Kronenberger et al., 2021) was completed by one parent in a separate room during testing.

Measures

Demographics, Hearing History, and Nonverbal Intelligence

Demographic information and hearing history were obtained by review of the medical record (when available) and/or parent interview. Age, gender, and household income, as measured on a 1 (< $20,000 per year) to 8 (≥ $150,000 per year) scale, were collected for both hearing groups. For CI users, age at onset of deafness, proxied by the age at which deafness was detected (n = 49); duration of deafness (n = 49); unaided preimplant PTA, measured at 500, 1000, and 2000 Hz in the better hearing ear (n = 36); age at implantation (n = 49); duration of CI use (n = 49); unilateral versus bilateral CIs (n = 51); and most recent CI-aided PTA (n = 47) were also collected. Nonverbal IQ was measured using the Leiter-3 CA (Roid et al., 2013), in which subjects are presented with a series of (pictured) objects or designs and instructed to identify an analogically or conceptually related picture. Performance on the Leiter-3 CA was expressed as age-based scaled score (with M = 10 and SD = 3), relative to a large, representative, normative sample.

Functional Hearing, Rapid Phonological Coding in Speech Perception, and Language

Functional hearing quality, rapid phonological coding in speech perception, and vocabulary/language skills are among the most important hearing and speech-language outcomes following cochlear implantation (Geers et al., 2011; Kronenberger et al., 2021; Nittrouer et al., 2014), and we hypothesized that LNS error scores would be correlated with these outcomes. Measures of these fundamental outcome domains were selected in order to broadly sample hearing (functional hearing quality), rapid phonological coding in speech perception, language knowledge (vocabulary), and following spoken language directions.

Functional everyday hearing quality was measured with the QHS (Kronenberger et al., 2021), a 21-item parent-rated questionnaire. QHS items are divided into four domains/subscales (speech, localization, sounds, and effort), which are averaged to produce a total functional hearing quality score (α = .82). QHS total scores differentiate CI users and NH peers and are associated with speech perception and language functioning in CI users. QHS total score was used to assess parent-rated functional hearing quality (n = 43 NH and 47 CI).

Rapid phonological coding in speech perception was assessed with the Children's Test of Nonword Repetition (CNRep; Gathercole et al., 1994), where participants repeat nonwords of increasing difficulty, presented via audio recording. Nonword repetition includes components of speech perception, encoding, STM, and speech production and has been shown to be highly predictive of spoken language outcomes following cochlear implantation. Prior work shows that nonword repetition is related more highly to quality of phonological representations than to verbal STM (Nittrouer et al., 2014), supporting its conceptualization as a measure of rapid phonological coding in speech perception. CNRep performance was expressed as percentage of correctly repeated nonwords.

Single-word receptive vocabulary was measured by the Peabody Picture Vocabulary Test–Fourth Edition (PPVT-4; Dunn & Dunn, 2007), which has been widely used as a measure of vocabulary in CI outcome studies (Geers et al., 2011). Subjects choose one of four pictures per item that matches a word spoken by the examiner. Age-based standard scores (with M = 100 and SD = 15) derived from nationally representative norms were used in this study.

Finally, following spoken directions was evaluated using the Following Directions subtest of the Clinical Evaluation of Language Fundamentals–Fifth Edition (CELF-5 FD; Semel et al., 2013); the CELF-5 is a well-known language measure that has been used to assess clinical language outcomes following implantation (Geers, Strube, et al., 2011). For this subtest, the examiner provides a set of directions that the child follows by pointing to pictures. Thus, the CELF-5 FD subtest assesses the ability to understand/comprehend, remember/retain, and sequentially execute a set of directions spoken by the examiner (this set of abilities is subsequently referred to simply as “following spoken directions”). Age-based scaled scores derived from nationally representative norms were used in this study.

VWM

VWM was evaluated using the LNS subtest of the WISC-V (Wechsler, 2014), in which subjects are presented with a random, mixed series of letters and digits (hereafter, stimulus sequence) and instructed to recall the digits in ascending order, followed by the letters in alphabetical order (hereafter, correct response). Stimulus sequences are presented in sets of three equal-length sequences, starting with two sets of sequences of Length 2 (e.g., U-9, 2-Z, and F-7 in Set 1 and B-3, 8-X, and W-6 in Set 2), followed by three sets of sequences of Length 3. Sequence length thereafter increases by one with each set. Testing is continued until the participant fails all sequences in a given set or reaches the end of the test. Correct responses are counted to arrive at the LNS raw score (hereafter, total sequence score). To obtain a higher resolution view of LNS accuracy, we created two additional accuracy scores: letter sequence score (number of sequences in which the letters were recalled accurately and in correct order, regardless of accuracy of digits recalled) and digit sequence score (number of sequences in which the digits were recalled accurately and in correct order, regardless of accuracy of letters recalled). Thus, three accuracy scores were calculated for each subject: total sequence score, letter sequence score, and digit sequence score (see Table 1).

Table 1.

Letter–number sequencing accuracy and error scores.

Administration
Stimulus sequence	Series of letters and digits read aloud to subject by examiner	R-5-J-3-X
Correct response	Correct response to stimulus sequence, with numbers in ascending order, followed by letters in alphabetical order	3-5-J-R-X
Accuracy scores
Total sequence	Number of correct responses	3-5-J-R-X
Letter sequence	Number of responses in which all letters are accurately recalled and correctly ordered, regardless of any errors in digit recall or order	2-4- J-R-X
Digit sequence	Number of responses in which all digits are accurately recalled and correctly ordered, regardless of any errors in letter recall or order	3-5 -R-V-X
Error scores
Order error	Number of responses in which all letters and digits are accurately recalled (without omissions, additions, or substitutions), but where ≥ 1 letter or digit is out of order	3- R -5-J-X
Phonological error	Number of erroneously substituted letters or digits, where the substitute letter/digit and substituted letter/digit are phonologically confusable (see Table 2)	3-5-R-J- S
Nonphonological error	Number of erroneously substituted letters or digits, where the substitute letter/digit and substituted letter/digit are not phonologically confusable	3-5-R-J- V
Addition error	Number of extra letters or digits in response, when compared with stimulus sequence	3-5- 7 -J-R-X
Omission error	Number of missing letters or digits in response, when compared with stimulus sequence	3-__-J-R-X

Note. Examples are provided at right: For accuracy scores, underlined and italicized portions are examples of scored sequences (with 1 point per sequence); for error scores, underlined and italicized letters and digits are examples of scored errors (with 1 point per sequence for order errors and 1 point per letter or digit for other error types).

For error analysis, errors were coded into five categories: order errors, phonological substitution errors, nonphonological substitution errors, addition errors, and omission errors, which were defined so as to be both mutually exclusive and collectively exhaustive of all possible recall errors (see Table 1). Errors were automatically coded by a custom Excel program, according to the definitions in Table 1. Finally, error scores (order, phonological, nonphonological, addition, and omission error scores) were calculated by summing the number of LNS errors of each type across all administered stimulus sequences.

To distinguish phonological and nonphonological substitution errors, sets of phonologically similar letters were constructed based on shared vowel sounds and common phonological confusions identified in prior work, as shown in Table 2: The “A” set was made up of letters with the long “a” vowel sound (/eι/; Cole et al., 1983, 1990; Conrad, 1964), the “E” set was made up of letters with the long “e” vowel sound (/i/; Cole et al., 1983, 1990; Conrad, 1964), the “I” set was made up of letters with the long “i” vowel sound (/αɩ/; Cole et al., 1983; Conrad, 1964), the “U” set was made up of letters with the long “u” vowel sound (/yu/; Cole et al., 1990; Conrad, 1964), and the “Eh” set was made up of letters with the short “e” vowel sound (/ε/). Because “L” was not disproportionately confused for other “Eh” letters (F, M, N, S, and X) in previous studies of acoustic similarity (Cole et al., 1990; Conrad, 1964), it was not included in the “Eh” set. For similar reasons, the “Eh” set was divided into two distinct sets of phonologically confusable letters: the “M-N” and “F-S-X” sets. These final letter sets were supported by the acoustic similarity matrix generated by Conrad (1964), for which within-set confusions were 2.60, 1.86, 6.70, 2.16, 8.31, and 8.35 times as common as would be expected by random chance for our “A,” “E,” “I,” “U,” “M-N,” and “F-S-X” sets, respectively. Digits were added to acoustic similarity sets if they shared the same vowel sound (see Table 2). Letters and digits with > 1 syllable (W and 7) were not included in any set.

Table 2.

Phonologically confusable letter/digit sets.

Set	Letters/digits
A set	A, H, J, K, 8
E set	B, C, D, E, G, P, T, V, Z, 3
F-S-X set	F, S, X
M-N set	M, N
I set	I, Y, 5, 9
U set	Q, U, 2
None	L, O, R, W, 1, 4, 6, 7

Statistical Analysis

Descriptive statistics for audiological and hearing history variables were calculated for the CI sample, and descriptive statistics for chronological age, gender, family income, and nonverbal intelligence were calculated separately for both hearing groups (CI and NH). The average age, household income, and nonverbal intelligence of the NH and CI samples were compared using two-sided t tests. Age groups and gender were compared between hearing groups using chi-square tests.

LNS scores (Wechsler, 2014) for accuracy (total sequence, letter sequence, and digit sequence) and types of errors (order error, phonological error, nonphonological error, addition error, and omission error) were compared across hearing and age groups using 2 (CI vs. NH) × 3 (7–12 years, 13–17 years, 18–22 years) analyses of covariance (ANCOVAs) with nonverbal intelligence as a covariate. In order to test study hypotheses, the hearing group (CI vs. NH) comparisons of LNS error scores were of primary interest in these analyses, but the Age Group × Hearing Group interaction term provided an additional test of whether differences between hearing groups were different for different age groups. Consistent with study hypotheses, significant differences between hearing groups were expected on all LNS accuracy scores and on the LNS phonological error score. Alpha was set at .05 for all tests.

Regression analyses were performed to investigate the associations between LNS accuracy and error scores and criterion measures of functional hearing quality (QHS), rapid phonological coding in speech perception (CNRep), vocabulary (PPVT-4), and following spoken directions (CELF-5 FD), while accounting for hearing group, chronological age, and nonverbal intelligence. A hierarchical regression approach was used, with three blocks of variable entry. The first block of variables was composed of hearing group (with NH coded as 0 and CI coded as 1), chronological age, and nonverbal intelligence in order to account for variance explained by these variables prior to testing the LNS terms. The second block was composed of LNS total sequence score and all five error scores (digit sequence and letter sequence scores were excluded because of near-perfect correlation with total sequence score [r = .96 and .99, respectively]). The third block of variables consisted of the product term of hearing group and each LNS score, in order to test whether the association between LNS score and the criterion measure was different in the CI versus NH sample (i.e., interaction or moderator effect). This third block of variables was entered using a stepwise method, with only statistically significant variables (p < .05) entered into the analysis.

Results

Sample Characteristics

Descriptive characteristics for each sample are presented in Table 3. The majority (82%) of CI users in this study were congenitally deaf, with all CI users having onset of deafness prior to 18 months of age. Participants in the CI sample were, on average, 23.19 months of age at the time of implantation and had used their CIs for an average of 11.95 years. Thirty-nine (76%) of the CI sample had bilateral implants, with the remainder having a single implant. The NH and CI samples did not differ with respect to age, gender, or household income, but the NH sample scored higher than the CI sample on the measure of nonverbal intelligence (Leiter-3 CA scaled score; see Table 3). Despite this difference between samples in nonverbal intelligence, both samples scored greater than the normative mean scaled score of 10 on the Leiter-3 CA, t(52) = 5.81, p < .001 for the NH sample and t(50) = 2.98, p < .005 for the CI sample.

Table 3.

Sample demographics and hearing history.

Variable	Cochlear implant M (SD)	Normal hearing M (SD)	t
Chronological age (years)	13.77 (3.98)	13.72 (4.16)	0.06
Income level a	4.80 (2.20)	4.85 (1.98)	0.10
Nonverbal intelligence b	11.25 (3.01)	12.60 (3.26)	2.19*
Age of onset of deafness (months)	1.78 (4.27)
Duration of deafness (months)	21.41 (10.52)
Unaided preimplant PTAc	106.45 (9.50)
Age at implantation (months)	23.19 (10.33)
Duration of implant use (years)	11.95 (3.76)
Most recent implant-aided PTAc	21.06 (4.35)
Variable	Cochlear implant (n)	Normal hearing (n)	Chi-square
Gender (female/male)	27/24	29/24	0.03
Chronological age group
7–12 years	25	27	0.04
13–17 years	16	16
18–22 years	10	10
Age of onset of deafness
Birth	40
1–6 months	2
7–12 months	6
13–18 months	1
Device characteristics
Bilateral	39
Unilateral	12

Note. Degrees of freedom (df) for t tests = 102, except for income level (df = 101).

^a On a 1 (< $20,000 per year) to 8 (≥ $150,000 per year) scale.

^b By Leiter International Performance Scale–Third Edition Classification & Analogies subtest scaled score.

^c Pure-tone average (PTA) in dB HL at 500, 1000, and 2000 Hz in the better hearing ear.

^* p < .05.

^** p < .01.

^*** p < .001.

LNS Scores by Sample

LNS performance is summarized in Table 4. Results of 2 (hearing group: CI vs. NH) × 3 (age group: 7–12 years, 13–17 years, 18–22 years) ANCOVAs controlling for nonverbal intelligence showed significant main effects on all three LNS accuracy measures for hearing group (NH > CI) and age group (older > younger). Scores for order errors, nonphonological errors, addition errors, and omission errors did not significantly differ between NH and CI samples, but the CI sample committed significantly more phonological errors than the NH sample, F(1, 97) = 5.90, p < .02 (see Table 4). Despite this, the NH and CI samples did not differ in the total number of LNS errors, that is, sum of order, phonological, nonphonological, addition, and omission error scores: M (SD) = 10.11 (4.63) and 10.02 (4.12) for NH participants and CI users, respectively; ANCOVA F(1, 97) = 0.11, p < .75. Scores for phonological errors, nonphonological errors, addition errors, and omission errors did not significantly differ between age groups, but the age groups differed in the number of order errors, with the 7- to 12-year-old age group making more order errors than the 13- to 17-year-old age group, t(82) = 2.54, p < .05. The Age Group × Hearing Group interaction terms were not significant for any of the LNS accuracy or error scores, indicating that differences between hearing groups were similar across age groups (see Table 4).

Table 4.

Letter–number sequencing, functional hearing, rapid phonological coding, and language by hearing group.

Variable	CI M (SD)	NH M (SD)	Hearing group F(1, 97)	Age group F(2, 97)	Hearing Group × Age Group F(2, 97)
Letter–number sequencing accuracy scores
Total sequence	15.65 (4.06)	18.32 (3.36)	11.76***	15.20***	0.51
Letter sequence	15.84 (4.06)	18.60 (3.21)	11.26***	18.83***	0.10
Digit sequence	18.84 (4.73)	21.94 (4.23)	12.60***	14.68***	0.40
Letter–number sequencing error scores
Order	0.86 (1.04)	1.09 (1.56)	1.09	4.43*	0.36
Phonological	2.51 (1.88)	1.53 (1.38)	5.90*	1.50	0.11
Nonphonological	2.59 (1.83)	2.36 (1.82)	0.22	1.32	0.02
Addition	0.20 (0.49)	0.30 (0.61)	0.10	1.00	1.49
Omission	3.86 (3.92)	4.83 (4.03)	0.23	1.09	1.84
Functional hearing measure
QHS total score	6.19 (1.74)	8.88 (0.97)	59.14***	— a	— a
Speech-language measures
CNRep % correct	40.45 (17.17)	84.81 (10.92)	258.03***	0.14	6.46**
PPVT-4 standard	92.63 (19.54)	113.92 (15.74)	20.49***	1.24	2.89
CELF-5 FD scaled	7.96 (3.42)	11.55 (2.76)	21.12***	0.39	3.50*

Note. F values are for effects from 2 (hearing group) × 3 (age group) analyses of covariance with nonverbal intelligence as a covariate. Error degrees of freedom = 84 for QHS total score and 87 for CELF-5 FD scaled score because of missing data. CI = cochlear implant; NH = normal hearing; QHS = Quality of Hearing Scale; CNRep = Children's Test of Nonword Repetition; PPVT-4 standard = Peabody Picture Vocabulary Test–Fourth Edition, standard score; CELF-5 FD scaled = Following Directions subtest scaled score from the Clinical Evaluation of Language Fundamentals–Fifth Edition.

^a Main and interaction effects for age group are not reported for QHS scores because no parents of 18- to 22-year-olds in the NH group completed QHS scales.

^* p < .05.

^** p < .01.

^*** p < .001.

Hearing, Speech, Language, and LNS Scores

Compared with NH peers, CI users scored lower on all measures of hearing, speech, and language (see Table 4), and no age group main effects were found on those measures (see Table 4). Significant Age Group × Hearing Group interaction effects were found for CNRep scores and CELF-5 FD scores. However, in each of the three age groups, the NH sample scored better than the CI sample on CNRep (p < .001) and CELF-5 FD (p < .05) scores, with the exception of the 18- to 22-year-old group for CELF-5 FD (p < .15). Thus, despite significant Age Group × Hearing Group interaction effects for CNRep and CELF-5 FD, the NH sample consistently scored higher than the CI sample, independent of participant age.

Regression models were used to investigate associations between the hearing, speech, and language scores (QHS, CNRep, PPVT-4, and CELF-5 FD as criterion variables) and predictor variable blocks of sample characteristics (Block 1: hearing group [NH vs. CI], age, nonverbal intelligence [Leiter-3 CA scaled score]), LNS scores (Block 2: total sequence score, order error score, phonological error score, nonphonological error score, addition error score, and omission error score), and interactions of hearing group (NH coded 0; CI coded 1) and LNS error scores (Block 3). Block 1, composed of hearing group, age, and nonverbal intelligence, was significantly associated with QHS, CNRep, PPVT-4, and CELF-5 FD scores (accounting for 48%, 72%, 35%, and 44% of variance, respectively, p < .001; see Table 5). Within Block 1, hearing group was the only significant predictor of QHS and CNRep (with NH peers scoring higher than CI users). Hearing group and nonverbal intelligence, but not age, were significant predictors of PPVT-4 and CELF-5 FD; NH and higher nonverbal intelligence were associated with higher scores on those measures (see Table 5).

Table 5.

Regression models for functional hearing, rapid phonological coding, and language.

Variable	QHS a	CNRep b	PPVT-4 c	CELF-5 FD d
Block 1 – R 2	0.48***	0.72***	0.35***	0.44***
df	(3, 86)	(3, 100)	(3, 100)	(3, 90)
Hearing group	−0.67***	0.83***	0.47***	0.42***
Chronological age (years)	0.01	0.05	−0.11	−0.04
Nonverbal intelligence e	0.07	0.08	0.25**	0.44***
Block 2 – R 2	0.56***	0.79***	0.48***	0.64***
df	(9, 80)	(9, 94)	(9, 94)	(9, 84)
LNS total sequence score	0.39**	0.31***	0.42**	0.55***
Order errors	0.03	0.13*	0.06	0.11
Phonological errors	0.02	−0.12*	−0.07	−0.17*
Nonphonological errors	−0.12	0.08	0.04	0.06
Addition errors	0.05	0.04	0.08	0.02
Omission errors	−0.14	−0.08	0.06	0.01
Block 3 – R 2	—	0.80***	0.54***	—
df	—	(10, 93)	(10, 93)
Group × Omission	—	−0.21*	—	—
Group × Order	—	—	−0.34***	—

Note. Values for predictor variables are standardized regression (beta) coefficients; degrees of freedom (df) are lower for QHS and CELF-5 FD because of missing data. LNS = letter–number sequencing.

^a Quality of Hearing Scale total score.

^b Children's Test of Nonword Repetition % correct.

^c Standard score for the Peabody Picture Vocabulary Test–Fourth Edition.

^d Scaled score for the Following Directions subtest of the Clinical Evaluation of Language Fundamentals–Fifth Edition.

^e By Leiter International Performance Scale–Third Edition Classification & Analogies subtest scaled score.

^* p < .05.

^** p < .01.

^*** p < .001.

After accounting for Block 1 variables, Block 2 variables (LNS total sequence score and the five error scores) explained an additional 8%, 7%, 13%, and 20% of variance in QHS, CNRep, PPVT-4, and CELF-5 FD scores, respectively (p < .001). Higher LNS total sequence scores were associated with better performance on the QHS and PPVT-4, but no LNS error scores were associated with QHS or PPVT-4 scores. On the other hand, higher LNS total sequence scores and fewer LNS phonological errors were associated with better performance on the CNRep and CELF-5 FD tests (see Table 5).

Stepwise entry (p < .05) of hearing group (NH coded 0; CI coded 1) and LNS error score variables was used for Block 3. On the basis of the test of significance required for stepwise entry into the equation, this third block was only included in the regression models for CNRep and PPVT-4 scores. LNS omission errors were associated more strongly with lower CNRep scores in CI users than in NH peers, whereas LNS order errors were associated more strongly with lower PPVT-4 scores in CI users than in NH peers (see Table 5).

Discussion

The primary goals of this study were to investigate the elementary information-processing mechanisms underlying the observed VWM delays and variability in deaf youth with CIs and to better understand and explain associations between VWM and speech-language outcomes in this population. These mechanisms were assessed using a systematic error analysis of the LNS task, a cognitively intensive test of VWM, which has not been previously studied in CI users. As expected, CI users were found to perform significantly worse on the LNS test than NH peers. Consistent with the hypothesis that VWM delays are due, in part, to fragile, underspecified phonological representations of verbal information in STM, CI users made significantly more phonological errors than NH participants, whereas other types of LNS errors did not significantly differ between hearing groups. The propensity for making phonological errors on the LNS test was associated with poorer performance on measures of rapid phonological coding and following spoken directions but not vocabulary or hearing quality.

LNS Accuracy and Errors Across Hearing Groups

All three LNS accuracy scores were significantly lower in the CI sample than in the NH sample in ANCOVAs testing for effects of hearing group, age group, and their interaction while controlling for nonverbal intelligence. Importantly, no Age Group × Hearing Group interactions were significant, indicating that the differences between hearing groups on LNS accuracy scores were not associated with age, despite the broad age range of the sample. The finding of higher LNS scores in the NH sample compared with the CI sample is consistent with the frequently replicated finding of an overall VWM delay in children with CIs (AuBuchon et al., 2015; Burkholder & Pisoni, 2003; Fagan et al., 2007; Kronenberger et al., 2013; Nittrouer et al., 2013; Pisoni et al., 2011) and extends this finding to a much more cognitively demanding VWM task (Crowe, 2000). However, although poorer performance on VWM tasks is well documented in CI users, very little is known about the specific types of errors—and thus the underlying cognitive information-processing mechanisms—that contribute to this poor VWM performance.

Results from this study demonstrated that the number of phonological errors was significantly higher in the CI sample than in the NH sample, whereas other error types did not differ significantly between the two hearing groups. This result was found in ANCOVAs that also tested for main effects of age group and interaction effects of age group and hearing group while controlling for nonverbal intelligence. No Age Group × Hearing Group interactions were significant for any of the LNS error scores, indicating that the differences between the hearing groups on the number of phonological errors were not associated with age across the broad age range of the sample.

The finding of greater number of phonological errors in the CI sample compared with the NH sample is consistent with a model in which fragile, underspecified phonological representations of letters and digits in STM are responsible, at least in part, for poorer recall performance of CI users compared with NH peers on the LNS task (Pisoni et al., 2011). These fragile, underspecified phonological representations in the STM of CI users likely result from accumulated experience with the coarse-coded input of the CI (i.e., CIs do not provide highly robust encoding of phonological information in the speech signal, causing a loss of detailed information in the phonological and indexical channels of speech), which is reflected in the quality of stored phonological representations of speech in CI users. Additionally, limitations of auditory experience in CI users, both before and after implantation, may affect the development of robust phonological representations of speech signals in STM.

When interpreting these findings, it is important to consider how the LNS test was administered. In contrast to a previous LNS error analysis, where all items were administered to all participants in an NH sample (Mielicki et al., 2018), the LNS test in this study was administered in accordance with standard WISC-V instructions (Wechsler, 2014), such that the test was discontinued when the participant failed all three responses in a block. This administration procedure is based on evidence that participants typically fail to provide any further correct answers after a failed block. As a result, participants were administered different numbers of items depending on their performance: Participants who answered more items correctly were subsequently administered more, longer items and, therefore, primarily erred on more difficult, longer letter–number sequences. In contrast, participants who failed three items in an early (shorter) block made their errors on shorter letter–number sequences. In this way, each participant was pushed to the limit of their span. Hence, LNS error scores largely comprised those memory errors made at the individual limit or edge of each participant's immediate memory span, where errors represent early, primary influences on memory failure, as opposed to more catastrophic memory loss from a generally overwhelmed VWM system on sequences that far exceed memory span.

Additionally, because errors occurred at the limit of each individual's memory span, approximately the same number of errors were elicited (and scored) for each participant, allowing for direct comparisons of error scores across samples. We can therefore conclude that fragile/underspecified phonological representations of verbal stimuli in STM are the primary contributors to weaknesses in the VWM of CI users, since they were overrepresented in their early memory failures at the limit of their spans. It is possible that CI users could also show increased vulnerability to other types of LNS errors (e.g., order, addition, and omission) at span lengths far exceeding their abilities, had they been required to complete all items in the LNS task instead of terminating testing after three consecutively failed sequences in the same block.

LNS Scores, Hearing, and Speech-Language Outcomes

In hierarchical regressions, LNS accuracy (total sequence score) was significantly positively associated with measures of functional hearing (QHS), rapid phonological coding in speech perception (CNRep), vocabulary (PPVT-4), and following spoken directions (CELF-5), even after accounting for hearing group, age, and nonverbal intelligence. These findings are consistent with numerous previous studies that have demonstrated positive associations between measures of VWM and several foundational domains of speech and language functioning (Cleary et al., 2001; Fagan et al., 2007; Kronenberger et al., 2014). Additionally, the CNRep and CELF-5 FD tests involve a significant component of verbal STM, which is likely shared with LNS total sequence scores. However, this study extends earlier findings about VWM and language outcomes by showing that phonological errors, presumably resulting from fragile phonological representations of verbal stimuli in STM, were negatively associated with rapid phonological coding in speech perception (CNRep scores) as well as with following spoken directions (CELF-5 FD scores), even after accounting for LNS performance (total sequence score).

The significant independent association of phonological errors with CNRep score indicates a robust positive relationship between the speed/fluency of phonological coding during fast, automatic speech perception and the strength and robustness of phonological representations of verbal stimuli in VWM. Although the cross-sectional design of this study precludes definitive tests of causal direction, it is likely that this association is bidirectional. Robust phonological representations of words in VWM would likely facilitate phonological coding and speech perception by making phonological information more easily accessible during spoken language processing. Conversely, rapid/automatic phonological coding may strengthen the phonological representations of words in memory by providing stronger, more robust perceptual processing and sensory registration. Faster, more automatic verbal rehearsal speed has been associated with stronger VWM (Pisoni et al., 2011). Hence, rapid phonological processing in spoken language may relate directly to the underlying quality of phonological representations essential for robust VWM in CI users.

The association of LNS phonological errors with CELF-5 FD scores further indicates that the ability to form and maintain phonological representations in VWM is critically important in following complex directions, over and above the individual's actual memory span. Higher order language skills involving understanding and sequencing of spoken directions are particularly vulnerable to processing delays in CI users, even after accounting for vocabulary delays (Kronenberger & Pisoni, 2019). Furthermore, language skills that demand comprehension, memory, and sequencing are much more dependent on language processing fluency and speed than are more basic language skills such as vocabulary and fund of information (Kronenberger & Pisoni, 2019). Thus, encoding and maintenance of robust phonological representations in working memory may be obligatory for understanding, retaining, and following spoken directions, because the individual must hold more information in mind and process that information more quickly (requiring more robust phonological representations) when understanding and following directions compared with more basic vocabulary or knowledge-based tasks. In addition to these study results, this hypothesis is supported by prior findings of very strong associations between rapid phonological coding and higher order language skills (Kronenberger & Pisoni, 2019).

Regression results also produced two interaction effects, suggesting that some types of memory errors may be more strongly associated with speech-language outcomes in CI users than NH peers. The significant Hearing Group × Omission Error interaction term in the regression predicting CNRep (see Table 5), for example, indicates that CI users with more omission errors showed poorer rapid phonological coding skills on the nonword repetition test compared with NH peers. Rapid phonological coding is particularly difficult for CI users, who score much lower on measures of nonword repetition than NH peers (Smith et al., 2019). Thus, CI users may not fully encode all of the fine acoustic–phonetic details as rapidly as the sequence is presented, resulting in loss of memory for digits or letters in the sequence (omission errors).

A second hearing group interaction effect was found for order errors predicting PPVT-4 scores, such that more order errors were associated with lower PPVT-4 scores, particularly in the CI sample relative to NH peers. Poorer vocabulary may be associated with more order errors specifically in the CI sample because of differences in sequential processing abilities or memory strength (Conway et al., 2009). This finding requires further investigation.

Study Limitations

When interpreting the results of this study, it is important to consider several methodological characteristics and limitations. First, participants were from a relatively wide range of ages (7–22 years). While this age range encompasses shared components of childhood and adolescence, including ongoing brain development, schooling, and dependence on the family, the cognitive processes that individuals employ during spoken language and VWM tasks may vary significantly, in kind or proportion, over the course of this age range. As expected, older participants recalled more LNS sequences than younger participants, consistent with well-established improvements in working memory with age (Wechsler, 2014). However, no Age Group × Hearing Group interactions were found for LNS accuracy or error scores, demonstrating that hearing group differences found between groups on LNS scores were not associated with age, despite the broad age range of the group. Furthermore, age was controlled in regression models predicting hearing, speech, and language outcomes. Thus, these negative findings for age effects on LNS error score results suggest that study findings may apply broadly across several developmental stages during childhood and adolescence.

Second, the NH sample had significantly higher nonverbal intelligence than the CI sample. Because nonverbal intelligence, language, and working memory are positively associated (Schneider & Niklas, 2017), nonverbal intelligence represented an important potential confound to address in study analyses. As a result, nonverbal intelligence was statistically controlled in ANCOVAs and regression models. Furthermore, the average nonverbal intelligence score of both hearing groups exceeded the normative mean, indicating that both hearing groups were high functioning in terms of intellectual ability. Therefore, it is unlikely that unrecognized global intellectual delays contributed to results, and study results may be more applicable to higher functioning CI users.

Conclusions and Future Directions

This is the first study to use error analysis of the LNS test to investigate the information-processing mechanisms that underlie the observed VWM delays and variability in deaf youth with CIs. In addition to providing novel insights into the causes and effects of VWM delays in CI users, this study serves as a useful example of how detailed error analysis of the LNS test can be used to elucidate the elementary information-processing mechanisms that underlie clinically relevant VWM delays and the associations of these delays with hearing, speech, and language outcomes. Study results suggest that VWM delays and variability in youth with CIs are due, in part, to fragile and underspecified phonological representations of verbal information in STM, which result in phonological VWM errors and have downstream associations with rapid phonological coding and language skills such as following directions.

Future research should extend these study results using additional measures of speech, language, and VWM. For example, administration of measures of language comprehension such as the CELF-5 Understanding Spoken Paragraphs subtest may provide additional insight into associations of VWM error patterns and linguistic concept formation. Additionally, research is recommended to test associations of VWM error patterns with measures of speech recognition, sentence repetition, and sentence memory, which are considered among the gold standards of outcomes after implantation. Because quality of access to the spectrum of speech, as assessed by conventional speech recognition tests, affects the development of phonological representations of language, results of speech recognition tests may provide additional insights into associations between quality of phonological representations of language and VWM in CI users.

Future research should also examine associations of VWM errors and spoken language outcomes using other VWM measures and administration modalities. In this study, the LNS test was administered in its standard, live-voice spoken language format, where stimulus items were read aloud by the examiner, and responses were spoken by the participant. Therefore, we cannot entirely rule out the possibility that some LNS scores of CI users were adversely affected by audibility factors, which would be most likely to manifest as phonological errors if some phonemes (particularly consonants) were not perceived accurately. However, two factors suggest that the influences of audibility on LNS results do not fully account for study findings. First, the LNS test was administered in a quiet room in full view of the examiner's face by experienced ASHA-certified speech-language pathologists, who were instructed to speak with sufficient volume and clarity to ensure that participants could accurately hear the stimuli. As previous studies have shown that speech recognition by CI users is generally very good in these types of ideal listening conditions (Davidson et al., 2011), it is unlikely that speech perception was a major factor contributing to the LNS performance gap between CI users and NH peers. Second, phonological errors were not associated with functional hearing quality (see Table 5), and post hoc correlational analyses in the CI sample showed that phonological errors were not significantly correlated (p > .05) with age at onset of deafness (r = −.14), duration of deafness (r = .27), unaided preimplant PTA (r = −.06), age at implantation (r = .22), duration of CI use (r = −.27), and most recent CI-aided PTA (r = −.04). In order to further extend study findings while accounting for potential audibility influences, a study of VWM using a visual version of the LNS test is currently in progress.

Additionally, future investigation is recommended to evaluate interventions aimed specifically at strengthening phonological representations in memory in deaf youth with CIs as a potentially efficacious approach for enhancing VWM and spoken language outcomes. Training using different talkers, dialects, and other dimensions/attributes of speech offers the potential to improve the strength of phonological representations in memory, because efficient processing of indexical properties of speech (e.g., identity and emotional tone of talker/speech; Geers et al., 2013) is essential for robust encoding of speech signals in immediate processing and STM. For example, auditory training targeting indexical properties of speech may encourage storage of more robust phonological representations, as listeners are trained to attend to multiple attributes of the speech signal while encoding and retaining invariant properties across talker and tone. Similarly, training in speech recognition involving high-variability speech signals across talker and dialect may strengthen the ability to extract and robustly encode core phonological components of speech (Smith et al., 2019). Training with speech samples that systematically vary segmental intelligibility, semantic meaningfulness, background noise, and other challenging components of the speech signal may provide additional ways to improve the quality of phonological representations of words in memory (Pisoni et al., 1985). Currently, most auditory training programs target speech recognition outcomes, with less emphasis on strengthening phonological representations in VWM. In addition to assessing improvements in speech recognition after these interventions, researchers should investigate the extent to which these auditory training protocols improve phonological representations of words in VWM, using measures such as the LNS subtest.

Funding Statement

This work was supported by National Institute on Deafness and Other Communication Disorders Grant R01DC015257, awarded to William G. Kronenberger and David B. Pisoni.