How Hearing Loss and Cochlear Implantation Affect Verbal Working Memory: Evidence From Adolescents

Abstract

Purpose:

Verbal working memory is poorer for children with hearing loss than for peers with normal hearing (NH), even with cochlear implantation and early intervention. Poor verbal working memory can affect academic performance, especially in higher grades, making this deficit a significant problem. This study examined the stability of verbal working memory across middle childhood, tested working memory in adolescents with NH or cochlear implants (CIs), explored whether signal enhancement can improve verbal working memory, and tested two hypotheses proposed to explain the poor verbal working memory of children with hearing loss: (a) Diminished auditory experience directly affects executive functions, including working memory; (b) degraded auditory inputs inhibit children's abilities to recover the phonological structure needed for encoding verbal material into storage.

Design:

Fourteen-year-olds served as subjects: 55 with NH; 52 with CIs. Immediate serial recall tasks were used to assess working memory. Stimuli consisted of nonverbal, spatial stimuli and four kinds of verbal, acoustic stimuli: nonrhyming and rhyming words, and nonrhyming words with two kinds of signal enhancement: audiovisual and indexical. Analyses examined (a) stability of verbal working memory across middle childhood, (b) differences in verbal and nonverbal working memory, (c) effects of signal enhancement on recall, (d) phonological processing abilities, and (e) source of the diminished verbal working memory in adolescents with cochlear implants.

Results:

Verbal working memory remained stable across middle childhood. Adolescents across groups performed similarly for nonverbal stimuli, but those with CIs displayed poorer recall accuracy for verbal stimuli; signal enhancement did not improve recall. Poor phonological sensitivity largely accounted for the group effect.

Conclusions:

The central executive for working memory is not affected by hearing loss or cochlear implantation. Instead, the phonological deficit faced by adolescents with CIs denigrates the representation in storage and augmenting the signal does not help.

For the past several decades, evidence has accumulated revealing the benefits of early identification and intervention, including cochlear implantation, for infants born with severe-to-profound hearing loss. However, questions persist concerning whether those early interventions are sufficient in the long term. Research with a variety of children at risk for language or literacy weaknesses for reasons other than hearing loss shows that all too often challenges appear long after victory has been declared in the fight against language delays. There is even a name for this phenomenon: Scarborough (2001) labeled it “the fourth-grade problem,” because it is around this time that children at risk for language delays thought to have been resolved by early intervention can start showing chinks in their linguistic armor. This same situation has been observed for children with cochlear implants (CIs). Newborn hearing screening allows us to identify infants with hearing loss at or near birth. Auditory prostheses and early intervention are provided, and these children are sent off to mainstream educational settings at school age, with parents and clinicians confident those early supports in language learning will be sufficient to carry the children through school. Yet, some of these children encounter challenges in higher grades. The basis of these later occurring problems is not that the child's language is deteriorating. Rather, the newly encountered challenges likely arise because the linguistic demands of the school environment change in ways that do not conform to the language profiles of these children. Their skills may be well suited for the linguistic demands of the early school years, but academic language in higher grades is different. It uses specialized vocabulary that does not necessarily possess the phonotactic regularities of the child's everyday language (Baumann & Graves, 2010; Coxhead, 2000). Moreover, sentences used by teachers and classmates at higher grades become more complex, incorporating devices such as center-embedded clauses (e.g., “The dog that the boy chased ran away”). Such structures make communication in higher grades informationally dense, meaning more information is conveyed per unit of speech than in earlier grades (Nagy & Townsend, 2012; Uccelli et al., 2015). In addition, language becomes increasingly decontextualized from the child's everyday experiences, decreasing the child's ability to rely on those familiar experiences to aid comprehension (Schleppegrell, 2004). The broad goal of the study reported here was to explore the mechanisms that may explain one likely source of the challenges faced by adolescents with CIs as they advance into higher grades, namely, poor verbal working memory.

Verbal Working Memory and Hearing Loss

Verbal working memory is the capacity to hold verbal material in storage long enough to perform an operation on it. This capacity is especially important when the linguistic structure being processed is complex. The listener must be able to store verbal information, while new information is received so that early- and later-arriving information can be integrated. Although the ability to perform this task is essential to academic activities for all students, regardless of hearing status (Cowan, 2014), a strong working memory capacity is especially important when communication is hindered by poor signal quality (Classon et al., 2013; Pichora-Fuller et al., 2016; Rönnberg, 2003). Unfortunately, evidence across numerous studies shows that children with CIs have deficits in verbal working memory compared to same-age peers (AuBuchon et al., 2019; Bharadwaj et al., 2015; Burkholder & Pisoni, 2003; Davidson et al., 2019; Kronenberger et al., 2020; Pisoni & Cleary, 2003; Pisoni & Geers, 2000). Thus, the very children who could benefit most from strong verbal working memory are the ones with weaknesses in this function.

Sensorineural hearing loss imposes degradation on the acoustic speech signal, and use of a CI further degrades that signal. In addition, when an individual experiences a period of months or years with raised auditory thresholds, prior to receiving a CI, that lack of input is suspected of diminishing cognitive and linguistic functioning in an irreparable manner. Two primary hypotheses have been offered to explain how the lack of auditory experience and/or ongoing signal degradation associated with use of auditory prostheses—especially CIs—may affect cognitive functioning, specifically verbal working memory. The set of high-level cognitive functions, known as the executive functions, develop across childhood, at least partly based on sensory experience. The first hypothesis examined in this study stems from the idea that if sensory experience is limited early in life, development of these executive functions will be harmed. Working memory, whether verbal or nonverbal, is considered an executive function, so it is predicted to be diminished by that early auditory deprivation. Supporting this hypothesis is robust evidence showing that children with CIs perform more poorly than their peers with normal hearing (NH) on tasks measuring an array of executive functions (e.g., Beer et al., 2014; Castellanos et al., 2015; Figueras et al., 2008; Horn et al., 2004; Kronenberger et al., 2013, 2020). For example, Kronenberger et al. (2020) examined the development of language skills and executive functioning (controlled attention, inhibition, and working memory) across early childhood for children who had received CIs. Results revealed sustained deficits in executive functioning that could not be explained by concomitant deficits in language abilities.

A specific instantiation of the general hypothesis that sensory deprivation early in life can interfere with executive functioning is the auditory scaffolding hypothesis (Conway et al., 2009). This hypothesis suggests that deficits in auditory experience early in life hinder the acquisition of the sorts of sequencing skills underlying much of working memory. In support of this hypothesis, Conway and colleagues reviewed studies showing that children with CIs are poorer than children with NH at nonverbal sequencing tasks, such as tapping the thumb against each finger in succession and recalling the order of presentation of variously colored squares. The basis of this proposal is that the auditory modality is the sensory system most dependent on sequencing abilities because sensory input in this modality is fleeting. According to the auditory scaffolding hypothesis, early experience with sound supports the development of the cross-modal circuitry in the central nervous system responsible for serial-order representations in general. Consequently, working memory for both auditorily and visually presented material should be impacted by deficient auditory experience, just as Conway et al. described. In a more recent example of the effect, AuBuchon et al. (2015) reported that children with CIs were poorer at recalling the order of presentation of digits, even when they were presented visually on a computer monitor. These investigators, however, proposed that this finding is due to less efficient encoding of phonological structure in verbal short-term memory, rather than to a generalized sequencing deficit. Work by others seems to support this current interpretation with evidence showing that when the sequential information to be stored involves visual locations rather than verbal material, children with CIs perform similarly to children with NH (Terhune-Cotter et al., 2021). Thus, the long-term impact of a period of early auditory deprivation on the development of executive functioning, including working memory, remains uncertain. The current study was designed to explore this uncertainty.

The second hypothesis offered for why verbal working memory deficits are commonly observed in children with CIs focuses on the poor signal quality available to these children. The idea is that verbal working memory is highly dependent on the listener's capacity to recover phonological structure from the speech signal and use that structure to encode verbal material into temporary memory storage. Durable storage is dependent upon precise encoding. Because the acoustic structure specifying phonological representations is degraded for listeners with CIs, those representations are underspecified. Thus, encoding of verbal items in a memory buffer is imprecise. The working memory component responsible for this function is termed the phonological loop, according to multicomponent models of working memory; this component stores material upon which a central executive can operate (Baddeley, 2012; Baddeley & Hitch, 1974, 2019). Testing at younger ages with the same children who participated in this current study revealed both poorer verbal working memory and poorer phonological sensitivity. Moreover, strong relationships between phonological sensitivity and verbal working memory were observed (Nittrouer et al., 2013, 2017). The current study examined whether those findings and that relationship endured for these children as they progressed through childhood. According to this second hypothesis, deficits in working memory should be found when verbal stimuli are used, but not when nonverbal stimuli are used.

Current Study

To examine these hypotheses, two groups of adolescents were tested: one group with NH and typical language development and another group with congenital hearing loss who used CIs. For the most part, all children had been participating in a longitudinal study designed to explore the effects of congenital hearing loss on child development, including language acquisition. These adolescents were just about to enter high school when these data were collected, so this investigation shed light on the comparative working memory capacities of adolescents in these two groups, exactly when the need for these capacities was about to ramp up to handle the academic language they were about to encounter.

The experimental method employed to examine working memory in the current study was immediate serial recall. In this paradigm, the participant is presented with a string of items—words when verbal working memory is the focus of investigation. The participant must then recall the order of presentation of items. In this work, responses consisted of touching pictures on a computer monitor representing the items in the order heard. To minimize as much as possible the likelihood that verbal rehearsal would be used, the adolescents in this study were instructed (nicely) to keep their mouths closed. Verbal rehearsal, if invoked, would likely favor children with NH because it is possible that their speech production is clearer and more precisely represents word-internal phonological structure. Unlike free-recall paradigms, the critical metric in serial recall tasks is whether the items are recalled in the order presented; as such, this method provides a sensitive test of the auditory scaffolding hypothesis. Moreover, the same items can be used across trials with no loss of validity. The use of closed stimulus sets provides a benefit to testing participants with hearing loss because a check of recognition for the items can be done at the start of testing and serve as confirmation of that recognition for the entire test. Another benefit of using serial recall as the paradigm for this study is that different patterns of response are expected based on the extent to which a phonological code is used to store items in a memory buffer. The classic response pattern for a serial recall task when nonrhyming words are used is a U-shaped function, with more accurate recall at the starts of lists (the primacy effect) and at the ends of lists (the recency effect). The recency effect is severely reduced when items are not phonological in nature. For example, strings of environmental sounds show little-to-no recency effects (Nittrouer & Lowenstein, 2014, 2022). Similarly, the recency effect is reduced when phonological structure is made less salient, such as through the use rhyming words (e.g., Nittrouer et al., 2016). Yet another way to reduce the availability of phonological structure, and so diminish the recency effect, is to degrade the quality of the signal itself. This has been accomplished using noise-vocoded signals in which speech is replaced with a few channels of noise temporally modulated to replicate the amplitude structure of the original speech signal (Nittrouer & Lowenstein, 2014, 2022).

Further support is provided for the proposal that a recency effect is obtained in immediate serial recall only when items are phonological in nature with the use of a procedure in which a “suffix” is employed. In these experiments, an item is added to the end of the list—the suffix. That item does not need to be recalled. Results across studies using this procedure reveal that the recency effect is diminished if that suffix is phonological in nature (e.g., the spoken word “go”), but not if it is a nonspeech signal, such as a tone (de Gelder & Vroomen, 1992; Nairne & Crowder, 1982; Nairne & Walters, 1983; Rowe & Rowe, 1976). On the other hand, the recency effect is diminished by a silently articulated suffix (e.g., “go”) that is available to the subject only through lipreading, even when list items themselves are auditorily presented (Campbell & Dodd, 1980; Crowder, 1983; Salter & Colley, 1977). Orthographic presentations of these same words, however, do not serve to diminish the recency effect (Spoehr & Corin, 1978). Thus, it appears that the suffix effect is evoked only when that suffix is phonological in nature, meaning it presents articulatory structure. By extension, the robust findings of the suffix effect suggest that the list items the suffix is serving to eclipse must also be phonological (articulatory) in structure; otherwise, interference would not be observed.

The signals available through CIs face signal degradation similar in kind to that imposed by noise-vocoding. That means a specific consequence to working memory for children with CIs may be diminished recency effects, separate from overall reduction in recall accuracy, which would indicate that phonological structure is not utilized in storage to the same extent as it is for children with NH. Alternatively, it could be that children with CIs display evidence of utilizing phonological structure to encode words into a short-term memory buffer just as children with NH do. If those phonological representations are less precise, however, that encoding, and so durable storage, may still be impaired. In this care, overall accuracy of recall could be diminished. The current study measured recency effects to test these alternative predictions.

Five sets of materials were presented in the serial recall paradigm used here: one set of nonverbal, spatial (visual-only) stimuli and four sets of verbal (auditory-only) stimuli. Nonverbal, spatial stimuli were used specifically because they would not require processing in the auditory system. Working memory can be viewed as consisting of a central processor that receives input through one of two front ends: a phonological loop that operates primarily on verbal material presented auditorily or a visuospatial sketch pad that operates primarily on nonverbal signals presented visually (Baddeley & Hitch, 1974, 2019). To avoid the use of nonverbal materials that would nonetheless be presented through the auditory modality, a condition that would complicate interpretation, nonverbal stimuli were presented in the visual modality. In this way, the operation of the central processor could be examined independently of the need for auditory processing of any kind.

Three sets of verbal material consisted of nonrhyming words. For one set, these words were audio-only recordings from a single talker, which is the standard for similar experiments. In the other two sets, that signal was augmented in some manner. In one case, the signal was presented in an audiovisual format. It was hypothesized that the addition of the visual speech signal would enhance the availability of phonological information, an addition that could be especially useful for the adolescents with CIs. This hypothesis was based on earlier findings. In particular, Nittrouer and Lowenstein (2022) found that the addition of the visual speech signal to the noise-vocoded signal improved both overall accuracy of recall and specifically the recency effect to levels equivalent to those obtained for unprocessed (clear) speech signals presented in an auditory-only modality. Thus, articulatory structure provided by lipreading appeared to enhance phonological encoding in the memory buffer for these adult listeners. Perhaps, it was reasoned, it could do the same for the adolescents with CIs.

The second kind of signal augmentation involved the addition of indexical information, a goal accomplished by having each word recorded by a different talker. Previous investigations into the effects of using words produced by different talkers generally demonstrated a negative effect on recall (Goldinger et al., 1991; Martin et al., 1989; Mullennix et al., 1989), but that work typically mixed talkers and words across lists. Furthermore, outcomes have not been completely consistent. In particular, Goldinger et al. (1991) found that at slow rates of presentation, recall of words produced by multiple talkers was actually better than recall for single-talker lists. This finding led these authors to suggest that “ . . . voice information remains as an integral component of the memory representation of all items” (p. 158). Here, we restricted presentation of each word to a specific talker across lists. The hypothesis was that having talker-specific information associated with each word might support more distinctive representations, especially for the children with CIs. In the past, we have proposed that children with CIs may retain longer into childhood the sorts of holistic lexical representations attributed to younger children with NH (e.g., Nittrouer et al., 2014). Therefore, the prediction was proffered that this stimulus manipulation could enhance word representations, bolstering the uniqueness of each word and making encoding in a memory buffer more robust for listeners with diminished phonological sensitivity, such as children with CIs. Work by others has shown that listeners with CIs recognize indexical information (Tamati et al., 2021), and there was no reason to suspect these adolescents with CIs would be different.

Finally, rhyming words were presented. These words are inherently less phonologically distinct than nonrhyming words. Previous work has shown that serial recall for rhyming words is poorer, even for typical listeners with NH (Baddeley, 1966; Conrad & Hull, 1964; Guérard et al., 2012; Gupta et al., 2005; Nittrouer & Miller, 1999; Salame & Baddeley, 1986). As a result of the diminished distinctiveness, rhyming words should evoke reduced recency effects. These stimuli were recorded by a single talker (the same talker who produced the nonrhyming, single-talker words) with no augmentation.

Additional measures consisted of two phonological tasks. One task examined sensitivity to phonological structure in the speech signal, with minimal processing required. The other task examined processing of phonological structure per se. The inclusion of these two sorts of tasks made it possible to assess what specifically accounts for the challenges faced by adolescents with CIs in encoding verbal material into a memory buffer: problems in recognizing phonological structure in the first place or in processing that structure.

Another goal of the current study was to examine potential changes in verbal working memory across childhood, especially late elementary school to the start of high school. Although there are reliable data demonstrating poorer verbal working memory in children with CIs compared to their peers with NH, most studies have involved children in the elementary grades (i.e., sixth grade or below), when language demands—and so demands on verbal working memory—are lower. With few exceptions (e.g., Geers et al., 2013), the magnitude of this deficit at older ages has not been studied, so it is not known whether it increases or attenuates at these higher grades. It was deemed important to examine whether there is evidence of change in verbal working memory for children with CIs as they move into these higher grades, because improvements could bring these capacities closer to those of children with NH.

Overall, our understanding of this important executive function—verbal working memory—in children with congenital hearing loss who use CIs remains incomplete. It is critical that investigation of the problem is extended, especially as these children are entering higher grades. Academic language becomes more complex in these grades, and adequate working memory can facilitate the understanding of such language, even when the signal is not optimal.

Method

Participants

Data were collected from 107 adolescents: 55 with NH (27 male participants, 28 female participants) and 52 with severe-to-profound, congenital hearing loss who used CIs (23 male participants, 29 female male participants). Table 1 displays demographic data, including means, medians, and standard deviations. Only age at the time of testing showed a significant difference between groups, t(106) = 2.81, two-sided p = .006. The adolescents with CIs were older than the adolescents with NH by 3 months on average. This difference was not considered problematic because all adolescents had just completed the same academic grade, so they were equivalent in educational level. Furthermore, the difference was small in magnitude, and it would only benefit adolescents with CIs, if there was a benefit to be had. Because hypotheses were that adolescents with CIs would perform more poorly than adolescents with NH, any a priori benefit would not inappropriately bias outcomes.

Table 1.

Means, medians, and standard deviations for demographic and audiometric measures at eighth grade for adolescents with normal hearing (NH; n = 55) and adolescents with cochlear implants (CIs; n = 52).

Variable	NH			CIs
Measure	M	Mdn	SD	M	Mdn	SD
Age at time of testing	173	173	5	176	176	5
Socioeconomic status (out of 64)	36	36	14	33	35	11
Age at identification				6	3	7
Age at first implant				26	15	29
Age at second implant (n = 36)				55	46	37
Pre-implant auditory thresholds				100	100	16
Aided thresholds at testing				20	20	5

Note. Age is given in months. Pre-implant auditory thresholds are better-ear mean thresholds in dB hearing level for the three frequencies of 0.5, 1.0, and 2.0 kHz. Aided thresholds at the time of testing are mean thresholds for the same three frequencies.

Socioeconomic status (SES) was assessed using a two-factor scale on which occupation and highest educational attainment are ranked from 1 to 8, lowest to highest. These scores are multiplied together, and the product serves as the SES index. An SES index was computed for each parent, and the highest value was used as the family SES (Nittrouer & Burton, 2005). An SES score of 30 indicates that the parent had a 4-year university degree and a job commensurate with that level of education. Groups did not differ on overall SES. The lowest SES score for an adolescent in each of the NH and CI groups was 12, which indicates a parent who likely works in a retail-sales job or other service profession. The highest SES scores were 64 for these groups, which represents a position as a medical doctor, chief executive officer of a large corporation, or equivalent. Thus, a wide SES range was represented.

Audiologic measures for the adolescents with CIs are also shown on Table 1 and indicate that these adolescents generally received their first CIs at young ages. Thirty-six of these adolescents had two CIs, and three wore a hearing aid on the ear contralateral to their CI. All adolescents with CIs were fully mainstreamed, and none relied on sign language to communicate. Aided thresholds were measured at the time of eighth-grade testing, and these aided thresholds were in or close to the normal range. Adolescents with NH had their auditory thresholds measured at the time of testing. Thresholds for the octave frequencies from 0.25 to 8.0 kHz were better than or equal to 20 dB HL in both ears for all these adolescents.

The Leiter International Performance Scale–Revised (Leiter-R; Roid & Miller, 2002), was administered to assess nonverbal cognitive abilities. The Leiter-R is a completely nonverbal tool. During administration, neither the tester nor the subject talks, interacting instead by pantomime, not standardized signs. A composite score known as the “Brief IQ,” is derived from individual scores on four subtests: figure-ground perception, form completion, sequential order, and repeated patterns recognition. This set of measures is meant to capture the contributions of nonverbal cognition to fluid reasoning, including visual selective attention and memory. The last two subtests (Sequential Order and Repeated Patterns) provide especially sensitive metrics of the kind of sequencing abilities hypothesized to be disrupted by auditory deprivation early in life, according to the auditory scaffolding hypothesis (Conway et al., 2009). Scores for the Brief IQ and the four subtests are shown in Table 2. No adolescent had a score on the Leiter-R below 80, and scores given in Table 2 show that the groups were similar in performance. There was no group difference on the Brief IQ, nor on any one of the four subtests.

Table 2.

Means, medians, and standard deviations for scores on the Leiter International Performance Scale–Revised at eighth grade for adolescents with normal hearing (NH; n = 55) and adolescents with cochlear implants (CIs; n = 52).

Variable	NH			CIs
Measure	M	Mdn	SD	M	Mdn	SD
Leiter Brief IQ score	106	105	13	103	99	14
Figure ground	10.0	10.0	2.7	9.6	9.0	2.9
Form completion	11.2	12.0	2.2	10.7	10.5	2.1
Sequential order	10.4	11.0	3.3	9.8	9.0	3.2
Repeated patterns	11.6	13.0	2.2	10.8	12.0	2.6

Note. Scores for the Brief IQ are standardized scores (M = 100, SD = 15). Scores for the four subtests are scaled scores (M = 10, SD = 3).

All but nine of the adolescents with NH were participants in a longitudinal study and traveled to the laboratory from across the country each summer. University students did the testing each summer and spent the spring being trained on procedures. As part of that training, adolescents with NH from the local community came to the laboratory to participate. The nine adolescents with NH who were not in the longitudinal study were subjects in that practice testing. All nine were tested just prior to the start of testing with the longitudinal subjects, so student testers were adequately skilled by that time.

Subjects in the longitudinal study were enrolled when they were infants or toddlers (Nittrouer, 2010). To be included in the study from the start, children had to have had no medical problems that would be expected to delay language acquisition, other than hearing loss in the case of children with CIs. English was the only language spoken to the children at home. Parents had NH or hearing that was readily corrected to normal levels with hearing aids. These inclusionary criteria were applied to all adolescents participating as practice subjects, as well. For the adolescents with CIs, intervention up to school age had to have focused on spoken language, although it could have included sign language as additional support. All parents confirmed that their goals for their children with CIs were that they would be able to attend mainstream educational programs at school age without the need for sign language interpreters, and all these children were in such settings from kindergarten until the time of this testing, at the end of eighth grade.

Equipment

All testing took place in a sound-treated booth. Stimuli were stored on a computer server and presented through a Creative Labs Soundblaster soundcard, with a Roland MA-12C powered speaker placed 1 m in front of the adolescent at zero degrees azimuth. Custom-written software controlled the audio and visual presentation of the serial recall stimuli. Computer graphics (presented at 200 × 200 pixels) on a touchscreen monitor were used to represent each word, or to show the matrix in the case of spatial testing. Responses were collected by having adolescents touch the pictures, shown on the monitor, in the order recalled. The software kept track of scoring. For generation of verbal stimuli, audio samples were collected using a Shure MX195 lavalier microphone and a Marantz PMD661 solid state recorder. Video samples were recorded using a Sony HDR-XR550V video recorder and a Sony FM lapel microphone.

For the phonological tasks, stimuli were presented in audiovisual format with the same soundcard and speaker as that used for the serial recall task. For generation of these stimuli, video samples were recorded using the same equipment as for the working memory audiovisual stimuli. Subject participation was video-recorded using that same hardware.

Audiologic testing was done with a Grason-Stadler GSI 61 audiometer. Adolescents with CIs had aided thresholds tested in free field. Adolescents with NH had thresholds measured under TDH-39 headphones.

Stimuli and Materials

Working Memory

Four sets of verbal stimuli were created for testing. Three of these sets used six nonrhyming words, and one set used six rhyming words. The nonrhyming words were simple consonant–vowel–consonant nouns that could be readily represented by line drawings (e.g., soap). The rhyming words were consonant-[ᴂt] nouns that could also be readily represented by line drawings (e.g., hat). For the sets of stimuli labeled as nonrhyming words (meaning only one talker, presented in audio-only mode) and as rhyming words (also from that one talker, presented in audio-only mode), stimuli were recorded by a man with a Midwest dialect using a sampling rate of 22.05 kHz, 10-kHz low-pass filtering, and 16-bit digitization. For the set of audiovisual stimuli, the male talker was videotaped, using extreme care to make sure his head did not move between the recordings of samples. Five samples of each word were collected in random order for all three sets. The specific samples of each word used in the final set were selected to be as similar in intonation pattern as possible.

The same recording methods were used to create the set of words providing indexical information, except that recordings were made by each of six talkers: three female and three male. The three female talkers were a young girl, a young woman, and an older woman. All were native speakers of American English with Midwest dialects. The three male talkers were an adolescent boy with a Midwest dialect, a middle-aged man with a Midwest dialect, and a middle-aged man with a foreign accent. These selections were meant to maximize the uniqueness of each voice. One token of one word was used from each talker in the final set. Thus, every presentation of a specific word was from just one talker.

The spatial stimuli were generated during testing. A 2 × 3 matrix was displayed on the computer monitor, and cells lit up one at a time in random order. All cells lit up as white to avoid having color serve as an additional cue, and especially one that could be encoded with a verbal label (i.e., the color name).

In summary, five sets of stimuli were used: four sets of verbal (word) stimuli and one set of nonverbal (spatial) stimuli. The verbal stimuli consisted of (a) nonrhyming stimuli (nonrhyming words spoken by one talker; presented in audio mode), (b) audiovisual stimuli (nonrhyming words spoken by one talker, presented in audiovisual mode), (c) indexical stimuli (nonrhyming words, each spoken by a different talker, presented in audio mode), and (d) rhyming words (rhyming words spoken by one talker, presented in audio mode). The spatial stimuli consisted of squares on the computer monitor, lighting up in sequence.

Phonological Tasks

Two measures of phonological abilities were obtained: one measure of sensitivity to phonological structure and one measure of phonological processing. Both tasks have been used extensively in this laboratory (e.g., Nittrouer et al., 2022). For measuring sensitivity to word-internal phonological structure, a final consonant choice task was used. In this task, the adolescent was presented in audiovisual mode with a word that was repeated by the adolescent. Then, three words were presented, and the adolescent had to say which one ended in the same sound. For measuring phonological processing, a backward words task was used. In this task, the adolescent was presented with a word in audiovisual mode that the adolescent repeated. Then, the adolescent had to say the word with the order of phonemes reversed. All phoneme reversals resulted in real words (e.g., nips to spin). Both tasks had 48 items, and both were presented with custom-written software in MATLAB that also kept track of responses. The experimenter recorded responses on paper forms, as well, and all records were checked later using the video recordings of the subject responding.

Procedure

All procedures were approved by the institutional review board of the author's institution. The data reported here were collected as part of a larger test battery involving collection of several kinds of data. Adolescents traveled to the laboratory for 2 days of testing in the summer after their eighth-grade year at school. Parents gave informed consent before the start of testing, and adolescents provided informed, written assent.

Testing across the 2 days consisted of seven 1-hr modules. Adolescents came to the laboratory in groups of six and were tested in an alternating fashion: 1 hr of testing, then 1 hr of break time, and so forth. During one of the modules, all the working memory tasks were presented, in randomized orders across subjects. During another module, the phonological tasks were presented, again in randomized order. Training was provided prior to testing for both kinds of tasks. The Leiter-R testing was done during still another module.

Working Memory

For the working memory tasks, the adolescent sat in front of the computer monitor at a table with hands flat on the table. Training was provided at the start of each condition. Line-drawn pictures of each of the six words (2 × 2 in.) were shown at the top of the monitor. During this training, the adolescent heard each word (in whatever condition was being tested) one at a time with the same recorded sample as that used in testing. For the audiovisual condition, stimuli were shown in the middle of the monitor in a roughly 4 × 4 in. display. The adolescent needed to touch the picture representing each word upon hearing it, as shown in Figure 1. The adolescent had to identify all six stimuli correctly. This requirement ensured that all adolescents recognized all words. No adolescent had difficulty meeting this requirement on the first try. In the next and final training, the adolescent was instructed to touch the pictures in the order heard, as rapidly as possible while still ensuring accuracy. The experimenter demonstrated this procedure once, and the adolescent was given a chance to practice it. All adolescents demonstrated that they understood the procedure on the first try and then moved to testing.

Figure 1.

Illustration of the set up for the working memory task.

During testing, 10 trials with each stimulus set were presented. Each trial consisted of the six words in a different, randomly determined order, presented at a rate of one per second for all words presented in audio-only format and for the spatial stimuli. For the audiovisual stimuli, however, that rate was too fast to feel natural, so the onset-to-onset interval was changed to 1,300 ms. The adolescent completed testing with each set before moving to the next set. Adolescents were instructed to keep their hands on the table until all six words were presented. After all six words were presented, the pictures representing those words appeared at the top of the computer monitor. Adolescents were told not to say the words, but only to touch the pictures. As the adolescent touched each picture in the order of recall, that picture appeared in the middle of the monitor forming a row from left to right.

For the spatial stimuli, similar procedures were used, except that a matrix appeared on the monitor and squares lit up one at a time in sequence. The adolescent touched the squares in the order recalled.

Three scores were obtained for all five serial recall tasks and used in analyses. The first score consisted of the percentage of correct responses for each list position separately (i.e., correct recall out of 10 trials). The second score was the percentage of correct responses across the 10 trials, summing across list positions (i.e., correct recall out of 60). A measure of the recency effect was the third kind of score obtained. This was done with a custom-written MATLAB routine that identified the list position with the lowest score in each condition, typically the third or fourth position (out of the six positions). That low score was subtracted from the score for the final list position for that condition to obtain the measure of the recency effect.

Phonological Tasks

For the phonological tasks, the adolescent sat in front of a computer monitor on which the audiovisual stimuli were presented. Training consisted of two live-voice items during which the experimenter carefully explained the procedures. Then, six recorded practice items were presented. Test items were arranged in order of increasing difficulty. Testing was automatically discontinued by the software if six consecutive incorrect responses were recorded. Percent correct scores were used in analyses.

Results

Data Screening

The five sets of working memory scores and scores from the two phonological measures were assessed for homogeneity of variance and normal distribution. The working memory scores met criteria for homogeneity of variance and normal distribution. A slight lack of homogeneity of variance was observed in both measures of phonological processing: The ratio of variances for adolescents with CIs to adolescents with NH was 2.6 for the final consonant choice task and 1.4 for the backward words task. These differences were not considered serious threats to analyses because sample sizes were reasonably large and reasonably similar. Where normality of distribution is concerned, only outcomes from the final consonant choice task showed a skewed distribution. To handle this concern, an arcsine transformation was applied and those transformed values were used in further analyses.

Developmental Effects and Reliability

Most of these adolescents tested in eighth grade had also been tested in fourth grade with some of the same stimuli: 44 adolescents with NH and 46 adolescents with CIs. Using data collected in fourth grade, this first analysis was conducted to compare performance of these adolescents in fourth and in eighth grade, both as a way to examine how performance across these 4 years of development changed and to assess reliability. For this purpose, overall (across list positions) serial recall scores for the nonrhyming words were used. Figure 2 shows the relationship for scores collected at fourth grade (x-axis) and those collected at eighth grade (y-axis). The Pearson product–moment correlation coefficient between scores at these two ages was .602 (two-tailed p < .001). These results indicate that there was generally good reliability between scores obtained in fourth and eighth grades. Mean overall serial recall scores across groups at each age are shown in Table 3. Paired-samples t tests performed on these scores for each group separately revealed no significant differences across grades. These results indicate that verbal working memory generally did not change for these adolescents between fourth and eighth grades.

Figure 2.

Relationship of scores for serial recall accuracy of nonrhyming words from fourth grade (x-axis) and from eighth grade (y-axis).

Table 3.

Means, medians, and standard deviations for serial recall of nonrhyming words in fourth and eighth grade for adolescents with normal hearing (NH; n = 44) and adolescents with cochlear implants (CIs; n = 46).

Grade	NH			CIs
	M	Mdn	SD	M	Mdn	SD
Fourth grade	71	73	15	58	58	14
Eighth grade	70	69	18	58	58	17

Note. Scores are percent correct serial recall.

Accuracy of Serial Recall

Figure 3 shows position-specific outcomes for all five sets of working memory data for adolescents with NH and for adolescents with CIs. Adolescents in both groups were able to recall order of presentation for the spatial stimuli better than for any of the verbal stimuli by a wide margin. Order recall for the spatial stimuli is highly accurate and appears equivalent for both groups. There is little evidence of a recency effect for these stimuli. These results suggest that the central executive for working memory performed equally well for both groups of adolescents.

Figure 3.

Recall accuracy at each list position for each stimulus condition by adolescents with normal hearing and by adolescents with cochlear implants.

Turning attention to the verbal working memory stimuli, it is apparent from Figure 3 that the adolescents with CIs performed more poorly overall than the adolescents with NH. Nonetheless, the pattern of response was similar across the two groups, with primacy and recency effects evident. Adolescents in both groups were least accurate at recalling serial order for the rhyming stimuli, as predicted. The effects of the two kinds of signal enhancement (audiovisual and indexical), however, appear to have varied across groups. For adolescents with NH, indexical information appears to have had no effect: The function for these stimuli is very close to the function for the nonrhyming stimuli. Visual information, on the other hand, appears to have improved recall accuracy, especially toward the ends of lists. Results for the adolescents with CIs seem to show that visual information had no effect on recall, compared to recall for the nonrhyming words, but indexical information appears to have diminished recall accuracy.

To test the veracity of these impressions gathered by visual inspection of Figure 3, a three-way, repeated-measures analysis of variance (ANOVA) was performed on the data for the four sets of verbal working memory stimuli. Data for the spatial stimuli were not included because those outcomes were so distinct compared to outcomes for the verbal stimuli. Table 4 shows the results of the ANOVA. All three main effects were significant. The significant condition effect likely reflects the fact that adolescents in both groups performed more poorly for rhyming stimuli. The significant position effect reflects the fact that recall accuracy varied across list position, with higher accuracy at the start (primacy) and end (recency) of lists. The significant group effect supports the conclusion that the adolescents with NH performed more accurately than the adolescents with CIs.

Table 4.

Results of the three-way analysis of variance performed on outcomes at eighth grade for the verbal working memory stimuli.

Effect	df	F	p	η2
Main effects
Condition	3, 315	115.22	< .001	.523
Position	5, 525	219.50	< .001	.676
Group	1, 105	22.33	< .001	.175
Two-way interactions
Group × Condition	3, 315	3.51	.016	.032
Group × Position	5, 525	1.31	.257	.012
Condition × Position	15, 1575	10.56	< .001	.091
Three-way interaction
Group × Condition × Position	15, 1575	1.53	.088	.014

Note. Condition included nonrhyming, audiovisual, indexical, and rhyming stimuli. df = degrees of freedom.

When it comes to two-way interactions, the Group × Condition interaction was significant, likely a reflection of the differences in effects of the two kinds of signal enhancement for these two groups of adolescents. The two-way interaction of Condition × Position was also significant, probably due to an apparent diminishment in the recency effect for the rhyming words. The three-way interaction of Group × Condition × Position was not significant, indicating that the effect of condition on patterns of recall across positions was similar for both groups of adolescents. Because there was no three-way interaction and the recency effect was similar across groups (i.e., there was not a significant Group × Position effect), mean overall scores for recall accuracy across list positions were used in further analyses. These mean scores are shown on Table 5, along with mean scores for the spatial stimuli. The number of subjects for the spatial stimuli for adolescents with CIs is 51, rather than 52, because data were lost for one adolescent due to experimenter error in saving. Table 5 also presents the results of between-groups t tests performed on outcomes for each set of stimuli. These outcomes reveal that the adolescents with CIs performed more poorly than the adolescents with NH in all conditions, except for the spatial condition. The effect size was largest for indexical stimuli, likely because the adolescents with CIs were hindered in recovery of phonological structure in this condition. The next highest effect size was for the audiovisual stimuli; in this case, that is likely due to the fact that the adolescents with NH benefited in this condition, whereas the adolescents with CIs did not. Effect sizes were similar for the nonrhyming and rhyming conditions.

Table 5.

Means, medians, and standard deviations for overall recall accuracy in eighth grade for adolescents with normal hearing (NH; n = 55) and for adolescents with cochlear implants (CIs; n = 52), along with outcomes of t tests performed on each set of data.

Variable	NH			CIs			t	p	Hedges's g
	M	Mdn	SD	M	Mdn	SD
Nonrhyming	71	70	17	59	58	18	3.578	< .001*	0.69
Audiovisual	75	78	18	60	59	20	4.273	< .001*	0.82
Indexical	70	70	15	52	53	16	5.773	< .001*	1.11
Rhyming	51	47	18	41	39	14	3.206	.002*	0.62
Spatial	88	90	9	85	85	10	1.344	.182	0.26

Note. Scores are percent correct serial recall. n for adolescents with cochlear implants for spatial stimuli = 51. Degrees of freedom are 105 for all data sets, except for spatial stimuli where degrees of freedom are 104. All p values are for two-tailed tests. Comparisons that remain significant at p = .05 when a Bonferroni correction is applied are marked with asterisks.

To examine differences across conditions for each group of subjects, separate paired t tests were conducted. Outcomes are shown in Table 6. These outcomes largely match impressions gleaned from Figure 3. For the adolescents with NH, results for the audiovisual stimuli were apparently just enough higher than those for nonrhyming stimuli and results for the indexical stimuli were just enough lower that outcomes were significantly different for the audiovisual and indexical stimuli. For the adolescents with CIs, all comparisons were statistically significant, except for the comparison of outcomes for the nonrhyming and audiovisual stimuli. For both groups of adolescents, the largest effects were observed when any outcomes for stimuli using the nonrhyming words (nonrhyming, audiovisual, and indexical) were compared to outcomes for the rhyming words. These results indicate that all adolescents were using phonological structure to encode words into a working memory store: When phonological structure is similar across words (e.g., they rhyme), verbal working memory is diminished. Furthermore, these results suggest that adolescents in neither group benefited significantly from the presence of additional information, either visual or indexical; in fact, the presence of indexical information interfered with storage of verbal material for adolescents with CIs.

Table 6.

Outcomes of paired-samples t tests for adolescents with normal hearing (NH; n = 55) and adolescents with cochlear implants (CIs; n = 52).

Contrast	df	t	p	Hedges's g
NH
Nonrhyming vs. audiovisual	54	−2.322	.024	−0.31
Nonrhyming vs. indexical	54	1.033	.306	0.14
Nonrhyming vs. rhyming	54	10.327	< .001*	1.38
Audiovisual vs. indexical	54	3.332	.002*	0.45
Audiovisual vs. rhyming	54	11.549	< .001*	1.55
Indexical vs. rhyming	54	9.624	< .001*	1.29
CIs
Nonrhyming vs. audiovisual	51	−0.453	.653	−0.06
Nonrhyming vs. indexical	51	4.172	< .001*	0.57
Nonrhyming vs. rhyming	51	9.000	< .001*	1.24
Audiovisual vs. indexical	51	4.547	< .001*	0.62
Audiovisual vs. rhyming	51	8.744	< .001*	1.20
Indexical vs. rhyming	51	7.180	< .001*	0.99

Note. All p values are for two-tailed tests. Comparisons that remain significant at p = .05 when a Bonferroni correction is applied are marked with asterisks. df = degrees of freedom.

Outcomes for adolescents with CIs were examined separately to evaluate how factors related to cochlear implantation might be affecting their working memory abilities. First, age of receiving a CI was correlated with each measure of working memory, both the four sets of verbal stimuli and the spatial stimuli. None of these correlations was significant. Next, the effects of having one or two CIs were examined using t tests. Again, none of these outcomes was significant.

Recency Effect

The recency effect for each set of stimuli was examined as an additional way of assessing the extent to which these adolescents used phonological structure for encoding items into a working memory buffer. Table 7 shows mean recency effect scores for each group, and Table 8 shows paired t tests comparing results across conditions for each group separately. Independent-samples t tests were also performed with these data, but none of those tests revealed a difference in the magnitude of the recency effect for adolescents with NH and adolescents with CIs; overall, the recency effect was similar across groups. Additionally, the recency effect was largest for the nonrhyming word lists, but neither of the enhanced signals affected the recency effect: Effects were similar for the nonrhyming, audiovisual, and indexical stimuli. That finding indicates that neither kind of enhanced signal influenced the extent to which phonological structure was used to encode items into memory, even though it appears from Figure 3 that visual information may have increased the recency effect for adolescents with NH. Rhyming words showed the smallest recency effect of the word lists: Significant differences were generally found for each of the nonrhyming, audiovisual, and indexical stimuli compared to the rhyming stimuli. That finding likely reflects the fact that phonological structure was least available in the rhyming stimuli. Overall, however, the recency effect was smallest for the spatial stimuli, probably reflecting the fact that there was no phonological structure at all in the spatial stimuli.

Table 7.

Means, medians, and standard deviations for recency effects for adolescents with normal hearing (NH; n = 55) and for adolescents with cochlear implants (CIs; n = 52).

Variable	NH			CIs
	M	Mdn	SD	M	Mdn	SD
Nonrhyming	29	30	19	31	30	18
Audiovisual	30	30	21	33	30	25
Indexical	28	20	17	34	30	19
Rhyming	23	20	15	22	20	15
Spatial	12	10	12	11	10	10

Note. Scores are percent correct difference scores. n for adolescents with cochlear implants for spatial stimuli = 51.

Table 8.

Outcomes of paired-samples t tests for recency effects for adolescents with normal hearing (NH; n = 55) and adolescents with cochlear implants (CIs; n = 52).

Contrast	df	t	p	Hedges's g
NH
Nonrhyming vs. audiovisual	54	−0.312	.756	−0.04
Nonrhyming vs. indexical	54	0.479	.634	0.06
Nonrhyming vs. rhyming	54	2.417	.019	0.32
Audiovisual vs. indexical	54	0.665	.509	0.09
Audiovisual vs. rhyming	54	2.547	.014	0.34
Indexical vs. rhyming	54	1.979	.053	0.27
Rhyming vs. spatial	54	3.934	< .001*	0.53
CIs
Nonrhyming vs. audiovisual	51	−0.620	.538	−0.09
Nonrhyming vs. indexical	51	−0.896	.375	−0.12
Nonrhyming vs. rhyming	51	2.212	.032	0.30
Audiovisual vs. indexical	51	−0.258	.797	−0.04
Audiovisual vs. rhyming	51	2.663	.010	0.37
Indexical vs. rhyming	51	4.078	< .001*	0.56
Rhyming vs. spatial	50	3.621	< .001*	0.50

Note. All p values are for two-tailed tests. Comparisons that remain significant at p = .05 when a Bonferroni correction is applied are marked with an asterisk. df = degrees of freedom.

Phonological Measures

Scores for the two phonological tasks are shown on Table 9, along with the results of independent-samples t tests. As expected, the adolescents with CIs showed deficits compared to their peers with NH.

Table 9.

Means, medians, and SDs for the phonological tasks for children with normal hearing (NH; n = 55) and for children with cochlear implants (CIs; n = 52), along with the results of independent-samples t tests.

Variable	NH			CIs
	M	Mdn	SD	M	Mdn	SD	t	p	Hedges's g
Final consonant choice	90	92	7	73	75	18	6.500	< .001*	1.25
Backward words	74	79	19	60	63	26	3.244	.002*	0.62

Note. Scores are percent correct. Arcsine transformations were used in the t test for final consonant choice. Degrees of freedom are 105 for all data sets. All p values are for two-tailed tests. Comparisons that remain significant at p = .05 when a Bonferroni correction is applied are marked with asterisks.

Explaining the Group Effect

The next set of analyses were intended to examine whether group differences in the central executive or the phonological loop accounted for the difference in verbal working memory between children with NH and children with CIs so well documented in the literature. Competing hypotheses exist, and one hypothesis is that the period of auditory deprivation early in life diminishes all functions considered to be executive functions, including working memory, regardless of whether those functions were specific to linguistic processes or not. According to this hypothesis, spatial working memory should be poorer for adolescents with CIs, indicating a diminishment in operation of the central executive component of verbal working memory. Of course, it has already been observed that no difference in nonverbal (spatial) working memory was found for these two groups of adolescents. Nonetheless, the question could be posed as to whether variability in the central executive across subjects explained differences between groups found for verbal material.

An alternative hypothesis is that the disproportionately large phonological deficits imposed on children with hearing loss by the degraded signal representation of CIs is responsible for the deficits in verbal working memory experienced by these children. This hypothesis puts the crux of the problem squarely on the phonological loop by suggesting that difficulties recognizing and utilizing phonological structure disrupt the storage of verbal material. These adolescents with CIs were found to have poorer phonological sensitivity and processing than their peers with NH; nonetheless, the question could be asked as to whether those phonological differences explained differences in verbal working memory.

To examine these alternatives, a series of four ANOVAs were performed using nonrhyming words as the dependent measure; because no strong effects of the enhanced signal structures were observed, it may be presumed that the nonrhyming word set represents outcomes for all nonrhyming stimuli (i.e., audiovisual and indexical, as well). First, an ANOVA with no covariate was done to see what the magnitude of the group effect, η², was. Next, scores for recall of the spatial stimuli and for the two sets of phonological measures were used as covariates to see how much the group effect decreased when each of these covariates was considered. If the group effect was diminished greatly or eliminated completely when the effect of one of these other measures was controlled, that finding would support the conclusion that whatever measure was responsible for the diminished effect explained the difference in verbal working memory between adolescents with NH and those with CIs.

Table 10 displays the outcomes of these analyses. Although scores for recall of spatial stimuli were related to scores for recall of the nonrhyming words, controlling for the spatial-stimuli scores had the smallest influence on the magnitude of the group effect, changing η² from .109 to .094. Both phonological processing measures had larger influences on the group effect. In fact, once the effect of the final consonant choice task—the measure of phonological sensitivity—was accounted for, the group effect was eliminated entirely. Controlling for the effect of the backward words task—the measure of phonological processing—greatly diminished the magnitude of the group effect, but did not entirely eliminate it. In this case, η² changed from .109 to .040. These outcomes support the claim that the large phonological deficit experienced by children with CIs explains the deficit in verbal working memory exhibited by this population of children.

Table 10.

Results of analysis of variance and analyses of covariance.

Covariate	df	F	p	η2
No covariate
Group effect	1, 105	12.805	< .001	.109
Recall for spatial stimuli
Covariate	1, 103	14.015	< .001	.120
Group effect	1, 103	10.717	.001	.094
Final consonant choice
Covariate	1, 104	18.522	< .001	.151
Group effect	1, 104	0.919	.340	.009
Backward words
Covariate	1, 104	39.533	< .001	.275
Group effect	1, 104	4.372	.039	.040

Note. df = degrees of freedom.

Discussion

Children with CIs in mainstream educational settings often face increased challenges when they reach higher grade levels in school, even if they managed to navigate earlier grades with little to no difficulty. These elevated challenges arise because the language of school becomes increasingly complex and more decontextualized from everyday experiences. To handle these challenges, adolescents with CIs need to rely on executive functions even more than adolescents with NH, as a means of compensating for any language deficits they may have. One executive function that is critical in these higher grades is verbal working memory. The language of lectures and other academic interactions in these higher grades is transmitted at faster rates, and information is transmitted in a denser fashion. Adolescents need to be able to recover phonological structure in a precise form and use it to store verbal material in working memory for processing. Research into verbal working memory on the part of children with CIs (mostly at ages younger than those examined here) has reliably demonstrated reduced capacity, compared to peers with NH. The purpose of the current study was to examine verbal working memory in adolescents with CIs as they were about to enter high school, see if it had changed (in particular, if it had improved) from younger ages, test two hypotheses regarding the mechanism responsible for any deficit, and explore whether augmenting the signal might help.

The first finding of this study was that verbal working memory capacity had not changed for these two groups of adolescents from fourth to eighth grade. This finding means that children with CIs continue to be at a disadvantage in terms of handling the language of school.

The two hypotheses examined as potential sources of deficit in verbal working memory were as follows: (a) The period of auditory deprivation experienced by these children early in life caused irreparable damage to all executive functioning, including working memory, and (b) the signal degradation imposed by CI processing means the phonological representations used for encoding verbal material into working memory are not as precise as those available to children with NH; thus, encoding is not as reliable.

Results of this investigation contradicted the first hypothesis, that children with CIs may have generally poorer executive functioning than children with NH. Adolescents in the two groups performed similarly on the measure of nonverbal intelligence, the Leiter-R. No differences were found in the Brief IQ score or in any of the four subtests. Furthermore, no group difference was observed in performance on the spatial working memory task. Thus, there were no differences in the specific executive function serving as the focus of this study, working memory, nor were there any differences in the executive functions measured by the Leiter-R.

The second hypothesis was well supported by the outcomes of this study. The adolescents with CIs demonstrated poorer sensitivity to phonological structure than their peers with NH, and when that phonological sensitivity was controlled, the group difference in accuracy of serial recall for verbal material was completely eliminated. Consequently, the model of verbal working memory deficit previously proposed (Nittrouer et al., 2013, 2017) receives support from this study: The poorer verbal working memory capacities so reliably reported for children with CIs appear to arise from difficulties storing verbal material in a durable fashion, because that storage is heavily reliant on the use of phonological structure and that structure is less precise for these children than for children with NH.

The final goal of this study was to examine whether supplementing the speech stimulus with another kind of signal structure—either visual or indexical—could bolster the encoding of verbal items in memory storage enough to compensate for the poor access to phonological structure available through audition with a single talker. Results did not support that proposal. Serial recall was no better for either group of adolescents when audiovisual signals, rather than audio-only signals, were used. That stimulus manipulation was meant to strengthen the availability of phonological structure through the provision of visual speech cues. In past work, it was found that when adults with NH are presented with clear auditory signals, they are able to recover phonological structure optimally and use it to encode verbal material into a memory buffer. The addition of visual cues to clear speech did not lead to improved recall accuracy in the way that it did when visual cues were added to vocoded speech; the addition of visual cues is likely redundant for adults with NH listening to clear speech (Nittrouer & Lowenstein, 2022). Results of this study indicate that the adolescents with CIs similarly failed to benefit from the addition of visual cues in their efforts to encode verbal material, a finding that was somewhat surprising because the auditory signals these listeners receive are degraded in a manner somewhat analogous to how vocoded speech is degraded for listeners with NH. In addition, when indexical information was provided in this study, the adolescents with CIs actually showed a small decrement in serial recall, indicating that the indexical information interfered with encoding and storage. It had been predicted this stimulus manipulation might enhance the distinctiveness of whole-word forms and would improve encoding and storage of words in a memory buffer, especially for the adolescents with CIs. The data did not support that prediction.

A new finding in these data was that although the adolescents with CIs were poorer at recovering phonological structure and using it to encode verbal material in a memory buffer, the psycholinguistic process employed for doing so was nonetheless the same for these adolescents as for those with NH: All subjects apparently recovered phonological structure from the speech stimuli—as well as they could—and used it to store words in a memory buffer. This conclusion arises from the results of the recency-effects analyses. These effects were of similar magnitude for the adolescents with CIs as for those with NH, in spite of the fact that overall accuracy was depressed for the adolescents with CIs. Thus, the adolescents with CIs were apparently utilizing the same process of relying on phonological structure for storage of material; they simply were not as successful at recovering clear structure from the speech signal.

Limitations and Future Directions

One limitation of the current study is the high recall accuracy for the spatial stimuli. Although mean scores were similar for both groups of adolescents, it is possible that an advantage may have been found for the adolescents with NH if the task had been more demanding, perhaps by using longer spans. That suggestion is based on Cleary et al.'s (2001) finding of poorer performance by 8- to 9-year-olds on a similar nonverbal memory task. In that experiment, there were just four cells and each one was a different color. As in the current study, these cells lit up individually in serial order. The dependent measure was the number of items that could be presented and result in accurate recall (i.e., the span). An accompanying condition, however, presented these items along with auditorily presented labels for each cell, which were the color words. This additional manipulation makes it difficult to compare across studies, because it is possible that the children with NH in the Cleary et al. study were using verbal labels to store presentation order, even when those labels were not presented. This interpretation receives particularly strong support due to the fact that the lights + verbal-labels condition always preceded the lights-alone condition. Furthermore, the finding in the current study that adolescents with CIs performed similarly to adolescents with NH on the Leiter-R lends additional support to the conclusion that the finding of similar nonverbal working memory was not spurious.

Another limitation of the current study is that there was no condition in which nonverbal stimuli were presented through the auditory modality. This design was implemented intentionally to avoid a potential confound. If a deficit was found for the adolescents with CIs using nonverbal, auditory signals, there would be no way of distinguishing whether that was because there is a general deficit in all working memory, or because acoustic signals are not processed adequately. Looking ahead, it would be of value to explicitly address that question. The multicomponent model of working memory serving as the basis of this study (e.g., Baddeley, 2012; Baddeley & Hitch, 1974, 2019) does not have a separate frontend for nonverbal auditory stimuli. Instead, it is generally agreed that such stimuli are processed by the phonological loop using global auditory representations (e.g., McKeown et al., 2011; Nees, 2016). Consequently, nonverbal, auditory stimuli could allow examination of the phonological loop without application of phonological structure itself—although subjects may apply linguistic labels even when the stimuli are not inherently linguistic, for example, using the words high and low to code tones that are higher or lower in frequency. Future studies will need to investigate further both working memory in children and adults with CIs for nonverbal, nonauditory stimuli that are more challenging than those used here and for nonverbal, auditory stimuli.

Neither did the current study explore serial recall for visual stimuli of phonological form. Nonetheless, it would be predicted that the adolescents with CIs would show similar deficits to those observed here for auditorily presented stimuli of phonological form, because earlier evidence has clearly revealed phonologically based deficits in tasks with visual stimuli. Children with CIs demonstrate reading and spelling deficits that can be attributed to poor phonological sensitivity (Werfel & Hendricks, 2023), and deficits have been observed when children with CIs have been asked to perform phonological awareness tasks with visual stimuli (Johnson & Goswami, 2010).

Finally, adolescents were restricted in their use of overt rehearsal in this study. This procedural manipulation means there is no way of assessing the potential contribution of rehearsal to verbal working memory for either group. Although overt rehearsal was prohibited, the adolescents may have been utilizing covert rehearsal and that rehearsal may have been more beneficial for those with NH precisely because their phonological representations are more precise. A future study will need to examine the effects of rehearsal on verbal working memory for children with CIs by having all participants engage in overt rehearsal.

Summary

The experiment described here was conducted to examine verbal working memory in adolescents with CIs using a serial recall task. Three primary objectives were addressed. First, verbal working memory across middle childhood was examined to see if those abilities might improve for children with CIs, bringing them closer to those of children with NH as they were about to begin high school. Results, however, revealed no improvement in serial recall between 10 and 14 years of age, for either group.

A second objective was to test two hypotheses regarding the source of the deficit in verbal working memory so reliably found for children with CIs. The first hypothesis was that auditory deprivation early in life, prior to receiving a CI, generally disrupts the development of executive functions for children born with hearing loss. In this study, no evidence was found to support this hypothesis, although it is acknowledged that mean scores for both groups were very high on the nonverbal working memory task, perhaps obscuring a group difference. The second hypothesis was that a primary phonological deficit interferes with recovery of phonological structure, thus degrading the representation of items to be stored in a memory buffer. Support was obtained for this hypothesis. In fact, it was found that when phonological sensitivity was controlled, the group difference in recall accuracy was no longer present.

Finally, the proposal was examined that verbal stimuli could be augmented for listeners with CIs as a way of enhancing representation in the memory store. To address this objective, two kinds of signal enhancements were applied: The visual speech signal was added to the auditory signal in one condition as a way to enhance phonological structure, and indexical information was added in another condition as a way to enhance global lexical structure. Neither kind of enhancement was successful in improving recall for these adolescents with CIs.

Overall, the results of this study provide insights into the challenges faced by so many children with CIs as they enter higher grades in school. The primary source of these challenges appears to rest with their poorly defined phonological representations, arising from the highly degraded auditory signals available through CIs. This situation suggests that we can mitigate the problems faced by these adolescents with accommodations—for example, preferential seating—and more intensive language intervention, but ultimately, solutions will depend on finding ways to provide better sensory information to individuals with severe-to-profound hearing loss.

Data Availability Statement

The data are available by request from the author: snittrouer@ufl.edu

Funding Statement

This work was supported by Grants R01 DC006237 and R01 DC015992 (awarded to Susan Nittrouer) from the National Institute on Deafness and Other Communication Disorders.