Nonword Repetition Assessing Auditory and Cognitive Abilities in Adult Cochlear Implant Users

Abstract

Purpose:

Phonological processing is essential for understanding spoken language and depends on auditory, cognitive, and linguistic skills. Although cochlear implants (CIs) can restore audibility in adults with sensorineural hearing loss, the speech recognition outcomes vary considerably. One factor previously found to underlie this variability is phonological processing skills, which may be additionally impaired by advancing age. This study examined phonological processing using a nonword repetition task (NWRT) with experienced adult CI users and normal-hearing (NH) peers. We also examined age-related changes in phonological processing and explored individual differences in auditory, cognitive, and linguistic factors that might predict NWRT variability among CI users.

Method:

Sixty-eight CI users and 44 NH adults older than 50 years completed an audiovisual NWRT and a battery of auditory, cognitive, and linguistic assessments, including measures of spectrotemporal resolution, reading efficiency, working memory, vocabulary, and nonverbal reasoning.

Results:

CI users demonstrated reduced accuracy on NWRT compared with NH peers. CI users also exhibited stronger age-related impairments in NWRT performance, suggesting compounding effects of hearing loss, degraded CI signals, and advanced age. Spectrotemporal resolution was the strongest contributor to NWRT performance among CI users, followed by reading efficiency, vocabulary size, and nonverbal reasoning.

Conclusions:

Our findings demonstrate that adult CI users experience challenges in phonological processing, with advancing age contributing to this decline. Performance on the NWRT reflects the combined influence of auditory (spectrotemporal resolution), cognitive (nonverbal reasoning), and linguistic (reading efficiency and vocabulary knowledge) factors and with spectrotemporal resolution emerging as the primary contributor.

Hearing loss in older adults is highly prevalent and is associated with challenges in speech communication (Humes & Roberts, 1990; Lin et al., 2011; Plomp, 1986) and reduced quality of life (Dalton et al., 2003). Cochlear implants (CIs) are well-established interventions for individuals with moderate to profound hearing loss, offering improved access to auditory information and leading to substantial gains in speech and language outcomes (Rasmussen et al., 2022; Wilson et al., 1991). However, adult CI users exhibit considerable variability in clinical outcomes, which are assessed using behavioral tests that measure open-set word and sentence recognition (Armstrong et al., 1997; Blamey et al., 2012; Boisvert et al., 2020; Schafer et al., 2011). This variability has been partly attributed to demographic and audiologic factors, such as duration of deafness, age at implantation (Blamey et al., 2012; Lazard et al., 2010), and individual surgical characteristics (Holden et al., 2013). However, these variables explain only a limited portion of the variance in the performance.

Auditory factors are strongly linked to speech recognition outcomes in CI users, largely because of spectrotemporally degraded signals delivered by implants, which limit access to fine-grained acoustic cues essential for speech perception (for a review, see Başkent et al., 2016). Spectrotemporal processing abilities are typically assessed using nonlinguistic auditory tasks, such as spectral ripple discrimination or modulation detection, and have shown strong correlations with speech recognition performance in both adult CI users and normal hearing (NH) listeners (Anderson et al., 2012; Henry et al., 2005; Litvak et al., 2007; Winn et al., 2016; Won et al., 2007). However, it remains unclear how much these low-level auditory discrimination abilities account for the variability observed in higher level speech outcomes among CI users.

Recent studies have increasingly explored the role of cognitive and linguistic mechanisms in explaining the variability in CI speech recognition outcomes (for a review, see Tamati et al., 2022). One specific area of cognitive and linguistic functioning that has been shown to be consistently related to speech recognition outcomes in adult CI users is the ability to process phonological information (Lyxell et al., 1998, 2003). Phonological processing is a broader cognitive–linguistic skill that additionally depends on working memory resources for storing and manipulating verbal information (Baddeley, 2001). A widely recognized speech recognition framework for understanding the interaction between auditory and cognitive mechanisms is the ease of language understanding (ELU) model (Rönnberg et al., 2013). According to this model, language processing is typically automatic when incoming auditory inputs match the stored phonological representations. However, when the signal is degraded, as it is when listening through a CI, mismatches occur, requiring more effortful top-down processing supported by executive resources such as working memory. In such cases, comprehension becomes slower and more effortful (Rönnberg et al., 2021, 2022), and individuals with stronger cognitive resources, particularly greater working memory capacity, may be better able to resolve these mismatches and improve speech recognition outcomes. As a group, adult CI users tend to perform poorer than their NH peers on tasks assessing phonological processing. For example, Moberly et al. (2016, 2017) demonstrated poorer performance by CI users than NH peers on tasks requiring the selection of words with the same starting or final sound as a target word or repetition of nonwords. This was true even when stimuli were presented in a combined audiovisual fashion (i.e., a talker's face showed speaking on a computer screen, while auditory stimuli were played in the sound field) to optimize sensory input. Additionally, adult CI users show poorer phonological processing in tasks using nonauditory visual stimuli, such as visual tasks of rhyme judgment and lexical decision (Andersson, 2002; Lyxell et al., 1998, 2003).

Measures of phonological processing have been examined as predictors of broader speech recognition outcomes, yet speech recognition itself may be influenced by a range of cognitive and linguistic abilities. For instance, reading efficiency and vocabulary size have been positively associated with speech recognition in experienced CI users, whereas nonword reading efficiency appears to be a stronger predictor in newly implanted individuals (Tamati et al., 2021). Additionally, nonverbal reasoning and working memory capacity have been shown to predict speech performance in both pre- and postoperative evaluations (Mattingly et al., 2018; Moberly et al., 2016; Moberly & Reed, 2019; Pisoni et al., 2017). Thus, investigating the relationships among these cognitive–linguistic measures and phonological processing may help explain the mechanisms underlying individual differences in speech recognition among CI users.

Another important factor influencing phonological processing and speech recognition is aging and the associated cognitive changes. Aging has been linked to physiological changes in the auditory pathways, as well as declines in key cognitive abilities, including slower processing speed, reduced working memory capacity, and diminished executive functioning, even in the absence of hearing loss (Hedden & Gabrieli, 2004; Pichora-Fuller & Singh, 2006; Ramadas et al., 2025; Roque et al., 2019; Tun et al., 2012). Numerous studies have shown that even older adults with normal audiograms perform less accurately in speech recognition, especially when processing fast or acoustically complex speech, which is partly attributed to age-related declines in temporal processing compared to younger listeners (Pichora-Fuller & Singh, 2006; Pichora-Fuller & Souza, 2003; Tremblay et al., 2003; Vaughan & Letowski, 1997). These situations often require greater reliance on top-down mechanisms such as attention and working memory (Füllgrabe et al., 2015; Nittrouer et al., 2016; Ramadas et al., 2025). Additionally, auditory deprivation has been identified as a potential risk factor for poorer speech recognition outcomes among CI users as absence of auditory stimulation can cause phonological representations in long-term memory to deteriorate over time (Lyxell et al., 2003). Although findings are mixed, evidence suggests that extended periods of auditory deprivation may have a negative influence on CI outcomes, especially if exceeding 15 years (Sorrentino et al., 2020). Older adult CI users may be particularly vulnerable to the cumulative effects of aging, peripheral hearing loss, limited auditory stimulation prior to implantation, and the reduced spectrotemporal resolution of the CI signal, all of which can place additional demands on cognitive and linguistic processing (Beyea et al., 2016; Moberly et al., 2017).

The Current Study

The current study examined phonological processing in adult CI users and NH peers using an audiovisual nonword repetition task (NWRT). Generally, the NWRT is a well-established test of phonological processing that engages a range of cognitive and linguistic mechanisms. This task requires participants to listen to and repeat novel nonwords that lack lexical meaning (e.g., ballop or altupatory). According to Gathercole's (2006) model, successful NWRT performance draws on multiple cognitive components including auditory discrimination, phonological encoding and storage, and speech motor planning and execution. NWRT outcomes have been consistently associated with phonological awareness, expressive language abilities, reading skills, and vocabulary achievement (Coady & Evans, 2008; Dillon et al., 2004; Dillon & Pisoni, 2006; Pisoni & Cleary, 2003; Schwob et al., 2021). These broader cognitive demands align with theoretical frameworks, such as the ELU model, which emphasizes the role of working memory in processing degraded input.

Research in children with CIs and specific language impairments (Dollaghan & Campbell, 1998; Ellis Weismer et al., 2000) has consistently shown that NWRT predicts language and word learning outcomes. Nittrouer et al. (2014) found that early NWRT performance in children with CIs predicted later language development and could help identify those at a higher risk for poor outcomes. Similarly, Carter et al. (2002) demonstrated that pediatric CI users were able to encode the prosodic envelope of nonwords despite limited access to detailed segmental information and that this prosodic knowledge was reflected in broader language and memory skills. Dillon et al. (2004) reported that NWRT accuracy in children with CIs is influenced by their speech perception abilities, verbal rehearsal speed, and speech production, with oral communication approaches showing particular benefits. Yet, despite it being widely used in pediatric populations, use of the NWRT in adult CI users is limited. For example, Moberly et al. (2017) reported that adult CI users performed significantly worse than did NH peers on an audiovisual NWRT, suggesting that they cannot access phonological representations effectively through the degraded auditory input delivered by a CI and that compensatory cognitive mechanisms were likely insufficient to overcome these limitations. At the group level, prelingually implanted children and postlingually implanted adults likely differ in their perceptual and processing abilities. Adults and children may exhibit distinct patterns of association due to developmental factors. Early-implanted children can benefit from increased neuroplasticity during the development of auditory pathways and language skills (Kral & Sharma, 2012), whereas older adult CI users are more likely to rely on established linguistic knowledge and cognitive resources (Gupta, 2003; Tamati et al., 2022). However, age-related declines in perceptual and cognitive functioning, combined with the degraded auditory input provided by a CI, may further constrain working memory and phonological processing, potentially contributing to reduced performance on tasks such as NWRT. Further examination of NWRT performance in the relatively understudied adult CI population may offer valuable insights into the mechanisms underlying speech recognition variability and inform targeted interventions.

The current study examined phonological processing abilities, assessed using an audiovisual NWRT, in experienced adult CI users and a group of similar-aged NH adults. More specifically, unlike previous studies that primarily addressed group-level differences, our study investigated individual differences in NWRT performance, by including auditory (spectrotemporal resolution), cognitive (working memory and nonverbal reasoning), and linguistic (reading efficiency and vocabulary knowledge) predictors in a relatively large sample of adult CI users. We hypothesized that, consistent with Moberly et al. (2017), CI users would demonstrate poorer NWRT performance than did NH peers, but we also hypothesized that individual differences in these auditory, cognitive, and linguistic measures would explain variability in NWRT performance in CI users. We further hypothesized that CI users would show a relatively stronger age-related decline in NWRT scores compared to their NH peers because of the combined effects of their experience of hearing loss and the spectrotemporally degraded signal provided by the CI.

Method

Participants

Sixty-eight adult CI users aged 50–87 years (M = 68.1, SD = 9.11) and a control group of 44 NH participants aged 50–81 years (M = 67.0, SD = 6.9) participated in this study. Both the CI and NH groups were sampled from similar age ranges to allow for the examination of aging-related effects across groups. The inclusion criteria for participation were chronological age above 18 years, prelingual or postlingual deafness for the CI group (defined as self-reported onset of hearing loss before or after age 12 years, respectively), native speaker of American English language, at least a high school diploma or equivalent, no diagnosed cognitive impairments, and at least 1 year of CI use (for the CI group). NH participants were tested as a control group, and similar age–grouped participants were included in the study, with the groups found not to differ significantly in mean age, t(110) = 0.68, p = .50. The inclusion criteria for NH participants were having NH, having no otologic concerns, and having no diagnosed neurological problems. NH was defined as a four-frequency pure-tone average (0.5, 1, 2, and 4 kHz) better than 25 dB HL in the better ear. For participants over 60 years of age, this criterion was slightly relaxed to 30 dB HL to account for age-related hearing changes; only three individuals exceeded the 25 dB HL threshold. Control participants were recruited through the Department of Otolaryngology and via ResearchMatch, a national research recruitment registry. Demographic questionnaires indicated that 51 CI participants had postlingual hearing loss (age at onset of hearing loss ranging from 12 to 68 years; M = 35.4 years, SD = 17.1 years); 14 participants had prelingual hearing loss (age at onset of hearing loss ranging from 0 to 12 years; M = 4.5 years, SD = 4.0 years). Three participants could not provide an approximate age for onset of hearing loss but confirmed that their hearing loss started after the age of 12 years and were identified as postlingual hearing loss. Among the CI participants, 30 were bimodal (i.e., CI + hearing aid in the nonimplanted ear), 20 had bilateral CI, 17 had unilateral CI users, and one CI user had an unspecified configuration due to missing information in the participants' demographics form. The CI participants were under standard audiological care at the program's associated clinic, and CI fitting and mapping were performed by a clinical audiologist during their regular visits. CI-aided thresholds were confirmed in the electronic medical record to be better than 30 dB HL across 0.5, 1, and 2 kHz in the sound field, indicating no concerns regarding audibility for the listening tasks in this study. Socioeconomic (SES) scores were calculated for participants based on their primary household income earners' occupational and educational levels, with higher scores indicating higher SES (Nittrouer & Burton, 2005). The CI participants had an SES of 6.0–64.0 (M = 26.1, SD = 14.6), and the NH participants had an SES of 9.0–64.0 (M = 36.3, SD = 14.2). To ensure the exclusion of any participants with significant cognitive impairment, an audiovisually presented Mini-Mental State Examination (MMSE; Folstein et al., 1975) was administered, and all participants demonstrated cognitive scores above the passing criteria (> 24 raw scores). The MMSE raw scores of CI users were between 26 and 30 (M = 28.6, SD = 1.3) and 26 and 30 for NH peers (M = 29.3, SD = 0.9). The two groups differed significantly for SES, t(92) = −3.54, p < .001, d = −0.70, with CI users demonstrating lower SES than NH peers, and for MMSE, t(110) = −3.28, p = .001, d = −0.59, with CI users demonstrating lower MMSE scores than NH peers.

Measures and Procedures

The participants completed the audiovisual NWRT and a battery of tasks designed to assess a range of auditory, cognitive, and linguistic skills. Auditory spectrotemporal processing was measured using the Spectral-Temporally Modulated Ripple Test (SMRT; Aronoff & Landsberger, 2013). Visual reading efficiency was assessed using the Test of Word Reading Efficiency–Second Edition (TOWRE-2; Torgesen et al., 1999). Working memory was measured using a visual forward digit span task (Wechsler, 1991). Word familiarity was evaluated using a visual WordFam task (Pisoni, 2007), and nonverbal reasoning was assessed using Raven's Progressive Matrices (Raven, 2000). These specific tests were selected based on our lab's previous findings that have demonstrated the relationship between these measures and CI speech perception outcomes in adult CI users (Tamati et al., 2021; Moberly et al., 2017). A detailed description of these tasks is provided below.

All tasks were completed in person and in one experimental session with breaks provided as needed. The total duration of the experimental session was approximately 60–90 min. The CI participants were tested with their own hearing devices in the best-aided bilateral condition, using their everyday listening configurations and settings to reflect typical listening experience. This study was approved by the local institutional review board. The study was approved by The Ohio State University Institutional Review Board (IRB 2015H0173). All participants provided informed, written consent and received $15 per hour for participation.

Phonological Processing (NWRT)

An audiovisual NWRT was conducted to assess phonological processing by requiring participants to repeat unfamiliar nonword stimuli presented in an audiovisual format on a computer screen and in a sound field. The NWRT stimuli used in this study were based on the materials developed by Dollaghan and Campbell (1998) and consisted of 40 nonwords (10 per syllable length, ranging from one to four syllables). In the current study, these stimuli were audiovisually recorded by a male speaker. Each syllable was equally stressed with a consistent fundamental frequency and amplitude. During the test, participants both saw and heard the talker to produce each nonword and were instructed to repeat it immediately. The sequence of nonwords began with 10 one-syllable words, followed by two-, three-, and four-syllable nonwords. Participants' responses were audio-recorded. Nonword repetition accuracy was scored at both the word and phoneme levels. Multiple trained scorers completed the scoring. To ensure consistency, 25% of the data were randomly selected and double-scored by another rater. Agreement across all scorer pairs ranged from 92% to 100%, indicating good interrater reliability. Thus, the scores from the primary scorer were used for data analysis. Nonwords were marked incorrectly if they were not repeated or articulated accurately. The total nonword scores ranged from 0% to 100%. Phonemes were considered incorrect only if they were omitted or substituted, whereas distortions (e.g., microphone or vocalization artifacts) were excluded. The total phoneme scores, calculated from a total of 343 phonemes, also ranged from 0% to 100%. The analyses focused on the percentages of correctly reproduced nonwords and phonemes across all syllable lengths.

Auditory Spectrotemporal Processing

Spectrotemporal processing, the ability to distinguish between different frequencies over time, was assessed using SMRT, which measures the spectral resolution by requiring listeners to detect ripples or variations in the frequency spectrum of sound. The SMRT does not rely on language comprehension, making it suitable for both NH and hearing-impaired participants. The task consisted of a three-interval forced-choice procedure. Two intervals present a reference stimulus with 20 ripples per octave (RPO), whereas the target stimulus initially has a 0.5 RPO. The target stimulus was modified using a 1-up/1-down adaptive procedure, with a step size of 0.2 RPO. The phase (φ) was randomly selected for each target and reference stimulus from one of four possible values: 0, π/2, π, and 3π/2. The test concludes after 10 reversals, and the thresholds are calculated based on the average of the last six reversals. In this study, stimuli were presented at 65 dB(A) through a speaker positioned in front of the listener at ear level, approximately 1 m from the head similar to the original study procedures (Aronoff & Landsberger, 2013). During the SMRT task, the participants were instructed to select the target stimulus using a computer interface. The performance was quantified in the RPO with higher thresholds, indicating better spectrotemporal resolution and that NH listeners typical range is between 4.5 and 8.5 RPO (Aronoff & Landsberger, 2013).

Visual Cognitive and Linguistic Measures

Both groups completed several visual nonauditory cognitive and linguistic tasks to assess reading efficiency, working memory, vocabulary size, and nonverbal reasoning. These tests were included because these cognitive constructs were previously reported to share underlying components with NWRT (Tamati et al., 2021; Wolf & Katzir-Cohen, 2001).

Reading Efficiency

Reading efficiency was measured using the TOWRE-2 (Torgesen et al., 1999). TOWRE-2 evaluates reading fluency and decoding ability by assessing the speed and accuracy of word and nonword reading. The test included two sets of items: a list of 108 words and a list of 66 nonwords. Participants were instructed to read aloud as many words and nonwords as possible within 45 s for each list. Two trained scorers used voice recordings to transcribe participants' responses. These scorers had previously reached 95% agreement with a trusted scorer on TOWRE-2 transcription. For each participant, one main and second trained scorer reviewed the TOWRE-2 responses and checked their scoring consistency. Only the word and nonword scores that showed at least 95% agreement between the two scorers were included in the analysis.

Working Memory

Working memory was assessed with a visual forward Digit Span Test based on the original auditory digit span task from the Wechsler Intelligence Scale for Children–Third Edition (WISCIII; Wechsler, 1991). In the forward Digit Span Test, participants were visually presented with sequences of digits on a computer screen and asked to recall them in the same order. Two sequences were presented at each span length before the sequence length increased. The test continued until the participant failed both sequences of a given length. Each correctly recalled sequence earned 1 point, and the total number of correctly recalled digits in the correct serial order was used as the score.

Vocabulary Size

Word Familiarity, as a marker of vocabulary size, was assessed using the WordFam Test (Pisoni, 2007), in which participants assessed their familiarity with a set of words. This test evaluates multiple dimensions of language ability, including lexical knowledge (understanding word meanings), semantic knowledge (relations and contexts of words), and lexical connectivity (how well words are associated with the mental lexicon). For this task, participants rated their familiarity with a set of 150 words on a scale ranging from 1 (completely unfamiliar) to 7 (very familiar). The words were categorized into high-, medium-, and low-familiarity levels based on normative data. The average score across all items in the WordFam Test serves as an indicator of vocabulary size.

Nonverbal Reasoning

The Ravens Progressive Matrices test was administered to assess nonverbal intelligence and abstract reasoning (Raven, 2000). The test consisted of a series of visual patterns or matrices with missing elements. Participants were required to identify the correct visual pattern that completes the pattern from several options. Scoring was based on the number of correct items completed in 10 min out of the total number of possible items.

Data Analysis

Descriptive statistics, including the mean, standard deviation, and range of NWRT total correct nonwords and phoneme scores, were calculated across different syllable lengths for each of the two groups. All analysis and visualizations were generated using R (Version 4.4.1; R Core Team, 2024).

To answer our first research question regarding group differences, we compared the NWRT scores of adults with CIs and those with NH and investigated the age effect for each group using generalized additive models (GAMs) with restricted maximum likelihood estimation, similar to previously published studies that used GAM models to evaluate age effects (Babaoğlu et al., 2024; Rachman et al., 2025). The advantage of using GAMs over traditional linear mixed-effects models is that they can capture both linear and nonlinear relationships between dependent and independent variables, where a nonlinear relationship is expected. This analysis was particularly suitable for examining the effects of age on the two groups. Separate GAMs were fitted for two dependent variables: NWRT total nonword scores and NWRT total phoneme scores and syllable lengths. Each model included a hearing group (CI vs. NH) as a parametric factor and a smooth term for age, modeled separately for each group using cubic regression splines (bs = “cs”) with up to five basis functions (k = 5). This approach allowed us to detect linear and nonlinear group-specific, age-related trends in the NWRT performance. To identify the age range at which CI and NH groups significantly differed, we used the plot_diff() function from the itsadug package, which computes the difference between group smooths and identifies time points with nonoverlapping 95% confidence intervals. All models were fitted using the MGCV (Version 1.8.42) and ggplot (Version 2.4.1) packages in R (Version 4.4.1, R Core Team, 2024), and model assumptions were checked via diagnostic plots and k.check() to confirm sufficient basis dimensions. We ran the model for NWRT percentage scores for total nonword and total phonemes using the following model to examine whether aging affects the two hearing groups differently.

$$\text{NWRTscore}~\text{Hearing}+\mathrm{s}\left(\mathrm{Age}\text{Years},\mathrm{by}=\text{Hearing}\right)$$ (1)

To further examine performance changes across syllable lengths, we fit a second GAM model including syllable length as an additional factor in each hearing group × syllable condition.

To answer our second research question regarding the relationship between individual auditory, cognitive, and linguistic factors and NWRT performance in CI users, we examined the total nonword accuracy scores in relation to each predictor variable. Total nonword accuracy was used instead of total phoneme scores to avoid a large number of comparisons in the analyses. Additionally, there was a strong and statistically significant correlation between total nonword and total phoneme accuracy scores (r = .77, p < .001). Finally, total nonword accuracy scores better represent higher level phonological processing demands and working memory capacity than total phoneme scores, which aligns with our aim of examining the relationship between NWRT performance and cognitive test performance. For this analysis, we conducted a series of GAMs, each including a single predictor, while controlling for age using a smoothing term. This approach allowed for flexible modeling of nonlinear age-related effects, while controlling for them when estimating the influence of each predictor. The β coefficient in each model reflects the unique contribution of the predictor to NWRT scores independent of age. The adjusted R² value quantifies how much variance in NWRT scores is explained by the model, considering both the predictor and the smooth term for age. Separate GAMs were fitted to each predictor: auditory spectral resolution (SMRT), reading efficiency (TOWRE), working memory (Digit Span), vocabulary size (WordFam), and nonverbal reasoning (Ravens). This approach allowed us to include all available data without excluding any CI participants who were unable to complete all the tasks due to their personal time constrains (see Table 1).

Table 1.

Results of generalized additive models predicting nonword repetition task (NWRT) total nonword accuracy scores across auditory, linguistic, and cognitive measures in cochlear implant users.

Predictor	β estimate	p (β)	p (age smooth)	Adj. R2	Deviance explained	N	Trends
SMRT	7.298	< .001**	.072	.335	36.7%	64	Strongest predictor; significant independent effect of spectral resolution; age not significant
TOWRE word	0.773	< .001**	.026*	.278	30.7%	68	Significant effect of word reading; age also contributes
WordFam	6.878	.004**	.0024**	.212	24.9%	62	Vocabulary and age both significantly related to NWRT
TOWRE nonword	0.344	.004**	.148	.183	21.1%	68	Moderate effect; independent of age
Ravens	1.427	.014*	.145	.173	21.2%	68	Modest effect of nonverbal reasoning; no age effect
Digit span	0.029	.842	.038*	.086	12.5%	68	No effect of working memory; age effect significant

Note. Predictors are ordered descending order of deviance explained. Adj, = adjusted; SMRT = Spectral-Temporally Modulated Ripple Test; TOWRE = Test of Word Reading Efficiency; WordFam = Word Familiarity Test.

^* p < .05.

^** p < .01.

Additionally, group comparisons between CI and NH participants auditory, cognitive, and linguistic measures were conducted using independent-samples t tests, and exploratory analyses were performed to examine potential audiological and demographic contributors to variability in NWRT performance in the CI group. Specifically, we assessed associations between total nonword scores and duration of deafness, age at first cochlear implantation, and age at onset of hearing loss.

Results

Group Differences in Auditory, Cognitive, and Linguistic Measures

Independent-samples t tests revealed significant group differences across auditory, cognitive, and linguistic measures (see Table 2). The largest group difference between CI and NH was observed on the SMRT test, with a mean scores of 2.06 versus 6.45 RPO, respectively. CI users also showed lower performance on TOWRE–Word (M = 73.10 vs. 77.11) and TOWRE–Nonword (M = 58.68 vs. 67.13), reflecting weaker reading and phonological decoding skills. WordFam scores were poorer in CI users (M = 4.66 vs. 5.27), indicating lower lexical familiarity. In cognitive measures, CI users scored below NH participants on Raven's Matrices (M = 10.12 vs. 12.98) and Digit Span (39.63 vs. 47.59). Overall, NH participants generally performed better, whereas CI users showed greater variability and overall reduced performance across auditory, cognitive, and linguistic measures.

Table 2.

Group comparisons between cochlear implant (CI) and normal hearing (NH) participants across auditory, cognitive, and linguistic measures.

Measure	CI N	CI, M (SD)	NH N	NH, M (SD)	t(df)	p	Cohen's d
SMRT (RPO)	64	2.06 (1.34)	43	6.45 (1.21)	−17.59 (97)	< .001	−3.40
TOWRE–Word (% words correct)	68	73.10 (11.27)	44	77.11 (9.38)	−2.04 (103)	.044	−0.38
TOWRE–Nonword (% words correct)	68	58.68 (19.94)	44	67.13 (15.68)	−2.50 (106)	.014	−0.46
WordFam (score)	62	4.66 (0.98)	42	5.27 (0.81)	−3.44 (98)	< .001	−0.66
Ravens (no. of items correct)	68	10.12 (4.44)	44	12.98 (5.85)	−2.77 (74)	.007	−0.57
Digit span (no. of items correct)	68	39.63 (16.11)	44	47.59 (17.44)	−2.43 (87)	.017	−0.48

Note. Values represent group means (M) with standard deviations (SD) in parenthesis. SMRT = Spectral-Temporally Modulated Ripple Test; RPO = ripples per octave; TOWRE = Test of Word Reading Efficiency.

Group Differences in NWRT Performance

For NWRT total nonword scores, the CI group had a mean score of 41.02% (SD = 19.69; range: 0–86.15), while the NH group averaged 87.18% (SD = 7.32; range: 68–100). For NWRT total phoneme accuracy scores, CI users had a mean accuracy of 80.36% (SD = 15.69; range: 16.91–92.42), whereas NH adults showed significantly higher performance with an M of 97.89% (SD = 1.64; range: 92.42–100). A larger effect size was seen when comparing CI users versus NH peers for total nonword scores (Cohen's d = −2.66) than for total phoneme scores (Cohen's d = −1.33), indicating a greater deficit in nonword repetition than phoneme accuracy when compared to NH participants.

Group Comparison for Hearing and Age Effects

The NWRT total nonword and total phoneme scores, as well as scores across syllable lengths for the two hearing groups, are presented in Figures 1 and 2, respectively. For total nonword scores, the GAM model revealed a statistically significant difference between groups, with NH participants scoring on average 45.5% points higher than CI users (p < .0001). A significant nonlinear effect of age was observed in the CI group (edf = 1.73, F = 2.98, p = .0012), indicating that increasing age was associated with a decline in NWRT total nonword scores performance, beginning around the age of 67.6 years (see Figure 1A). In contrast, no significant age-related changes were observed in the NH group (p = .94). The model explained 70.5% of the variance in total nonword scores (R² = .698).

Figure 1.

Nonword repetition task (NWRT) total correct nonword (A) and total correct phoneme (B) scores as a function of age for cochlear implant (CI; red) and normal-hearing (NH) adults (gray). Gray dots represent the NH (N = 44); red dots represent the CI (N = 68) participants' individual scores. The bold lines represent the predicted relationship between age and the percent correct scores. Shaded areas represent 95% confidence intervals. Bold red lines on the x-axis represent the estimated differences in NWRT scores between hearing groups indicating a significant group difference across all ages (50–87 years). The dashed blue line represents the estimated age of performance drop in the CI group.

Figure 2.

Individual nonword (top) and phoneme (bottom) percent correct scores of CI (N = 64) and NH (N = 44) participants shown for each syllable length (increasing left to right). Red dots represent the CI users, and gray dots represent the NH group. The solid line represents the regression lines, and the shaded area represents the 95% confidence intervals.

For NWRT total phoneme scores, NH participants demonstrated higher scores than CI users, with a mean group difference of 17.4% points (p < .0001). Age-related smooth function for the CI group was significant (edf = 1.01, F = 1.62, p = .020), whereas no age effect was observed in NH participants (p = .98). The model accounted for 36.5% of the variance in phoneme scores (R² = .353). Although the GAM results revealed a significant age effect on phoneme repetition scores in CI users, the slope of the age function did not show a statistically significant negative trend at any specific age, as observed in total nonword scores. This suggests that age influences the pattern of phoneme scores, but no clear age of onset is extracted. The NH group showed no significant age effect (edf = 0.00, p = .988).

In the analysis of different syllable lengths in total nonword scores, CI users performed below their NH peers across all syllable lengths, with the largest gap between groups observed at the four syllables (CI M = 30.0%, NH M = 82.0%). NH participants showed significantly higher scores across syllable length, with no additional interaction effects by syllable length. However, visual inspection of the participants revealed that NH total nonword mean scores also declined at four and five syllables (82.0% and 81.8%, respectively) when compared to scores for two and three syllables (91.5% and 91.1%). For phoneme scores, NH participants performed better than CI users, maintaining near-ceiling performance at all syllable lengths (see Figure 2).

Relation Between NWRT Total Nonword Scores and Auditory, Cognitive, and Linguistic Measures in the CI Group

Auditory spectrotemporal resolution (SMRT) was the strongest predictor of NWRT nonword performance (β = 7.30, p < .001), followed by word reading efficiency (TOWRE Word; β = 0.77, p < .001), vocabulary size (WordFam; β = 6.88, p = .004), nonword reading efficiency (TOWRE Nonword; β = 0.34, p = .004), and nonverbal reasoning (Ravens; β = 1.43, p = .014). In contrast, working memory (Digit Span) did not significantly predict the NWRT scores (β = 0.03, p = .84).

Demographic Factors in CI Group

Exploratory correlational analyses were conducted to examine the relationship between demographic factors and NWRT performance in CI users. The age at implantation among CI users ranged from 35.0 to 82.0 years (M = 61.9, SD = 12.3). The mean duration of deafness prior to implantation was 32.5 years (SD = 18.6; range: 1.0–72.0), and the mean age at onset of hearing loss was 29.1 years (SD = 20.2; range: 0–68). NWRT total nonword scores were only weakly inversely correlated with age at first cochlear implantation (r = −.26, p = .04), but not with duration of deafness (r = −.17, p = .17) or age at onset of hearing loss (r = −.03, p = .82).

Discussion

The primary aim of this study was to examine how phonological processing abilities are influenced by both hearing status and aging in experienced adult CI users compared with their NH peers. We hypothesized that audiovisual NWRT performance would be lower in adult CI users relative to their NH peers and that CI users would show a relatively stronger aging effect on performance. Consistent with our first hypothesis, our results revealed significant group differences in NWRT scores between CI and NH participants in both total nonword and phoneme accuracy. On average, NH listeners scored 45.5 and 17.4 points higher than CI listeners, respectively. Additionally, advancing age significantly impacted NWRT performance in the CI group, with an observed drop in performance on the total nonword scores that occurred at around 67.6 years. In contrast, NH participants showed near-ceiling NWRT performance, with no significant decline across age. These findings are consistent with prior studies reporting significantly poorer performance by postlingually deafened adult CI users on phonologically demanding tasks such as NWRT (Moberly et al., 2017; Tamati et al., 2021) and rhyme judgment (Lyxell et al., 1994, 1996, 2003). Additionally, the observed hearing group differences were supported by significant correlations between NWRT scores and auditory spectrotemporal resolution, as measured by the SMRT, which will be discussed in the following section.

The age-related decline in NWRT performance observed in the CI group supports our hypothesis that combined effects of hearing loss, signal degradation, and auditory–cognitive aging contribute to lower phonological processing scores. Although NH participants demonstrated near-ceiling performance across ages, CI users showed greater variability, with a decline in performance with older age. This pattern suggests compounding effects of aging, insufficient auditory stimulation, and degraded auditory input, which may place additional demands on top-down cognitive–linguistic mechanisms such as working memory and attention. Consistent with the previously discussed ELU model (Rönnberg et al., 2013), when the quality of auditory input is degraded and phonological representations are insufficiently robust, listeners depend more on explicit cognitive processes to resolve mismatches between input and stored representations.

Although we did not directly measure auditory deprivation, we examined proxies such as age at implantation, duration of deafness, and age at onset of hearing loss. Exploratory analyses indicated only a weak association for age at implantation with NWRT, with no significant relationship for the other measures, suggesting that deprivation-related proxies explained little variance compared to the other cognitive–linguistic factors. The observed association between lower NWRT performance and increasing age in the CI group may reflect the integration of the perceptual and cognitive mechanisms necessary for accurate phonological processing. Finally, it should be also noted that there is considerable heterogeneity among adult CI users, both in general and within the current study sample. Comprehensive information about each participant's hearing history, including exact duration and onset of hearing loss, residual hearing during their longitudinal experience of hearing loss, and hearing aid use (e.g., duration, daily use, and perceived benefit), is often difficult to obtain. Nonetheless, these relationships warrant further investigation in future longitudinal studies.

The NWRT in this study were presented audiovisually to provide visual support and reduce the confounding effects of limited audibility in CI users. Visual speech cues can enhance perception and reduce floor effects, particularly in individuals with severe hearing loss. This dual-modality format may help isolate auditory limitations from broader cognitive–linguistic factors, such as vocabulary knowledge, working memory, and executive function (Al-Salim et al., 2020; Nittrouer et al., 2014). In our study, CI users performed better on phoneme-level accuracy than on total nonword accuracy, likely because some individual phonemes are more perceptually salient and visually accessible. While visual input may have supported phoneme identification, repeating novel nonwords remained challenging for CI users, indicating that visual cues were not sufficient to fully compensate for degraded auditory input.

Additionally, evidence from vocoded speech studies demonstrates how cognitive–linguistic resources support speech recognition in NH participants when tested with degraded auditory inputs, providing a simulation of CI hearing. Even among adults with normal or near-NH, age has been shown to have a significant effect on vocoded speech recognition, with older adults generally performing worse than younger adults (Moberly et al., 2023). This age-related decline has been linked to reduced temporal processing abilities, which are critical for understanding spectrally degraded speech (Goupell et al., 2017; Sheldon et al., 2008). However, age-related declines in cognitive functions such as working memory, executive function, and processing speed may further compromise speech understanding under challenging conditions, such as background noise or degraded input (Gordon-Salant & Fitzgibbons, 1995, 1997; Pichora-Fuller et al., 1995; Ward et al., 2017). These findings reflect greater reliance on top-down mechanisms and underscore the dynamic interaction between hearing status, signal quality, and cognitive processing in phonological performance.

Although not statistically significant, visual inspection of individual data also revealed a decline in total nonword accuracy for longer syllables in the NH group. Syllable length influenced performance in both groups, particularly for longer nonwords (four and five syllables), with the effect being significant in the CI group, likely due to the increased phonological and short-term memory demands associated with longer stimuli. This pattern aligns with findings from Al-Salim et al. (2020), who used a similar audiovisual NWRT and reported that longer nonwords place greater demands on phonological working memory and motor planning in children with CIs compared to peers with mild hearing loss or NH. Similarly, Sadagopan and Smith (2013) examined speech motor performance as a function of nonword length and complexity in younger (M_age = 20.1 years) and older (M_age = 69.0 years) adults with NH. While younger adults achieved high phoneme-level and total nonword accuracy scores (≥ 95%), older adults showed a significant decline, particularly for four-syllable nonwords. As the nonword length increases, the processing demands grow, challenging the ability to maintain and reproduce the full nonword. The observed decline in performance likely reflects age-related limitations in working memory capacity when faced with increased linguistic complexity, with effects more pronounced in the CI group due to compounded auditory constraints.

The secondary aim was to investigate the effects of auditory, cognitive, and linguistic performance on phonological processing in order to better understand individual variability in NWRT performance among adult CI users. We hypothesized that NWRT performance in CI users would be predicted by their underlying auditory, cognitive, and linguistic abilities. The finding that auditory spectrotemporal resolution significantly predicts NWRT performance is consistent with prior studies that spectral ripple discrimination and spectrotemporal modulation sensitivity are strongly correlated with speech recognition outcomes in CI users (Anderson et al., 2012; Luo et al., 2020; Winn et al., 2016; Won et al., 2007). CIs are known to provide limited spectral resolution, which restricts the ability of the device to encode fine-grained acoustic cues (Başkent et al., 2016). This limitation may impair phoneme sequencing and the perception of transitions across syllables, particularly in longer nonwords. When spectral detail is reduced, listeners have less access to fine phonological cues, resulting in increased difficulty in speech understanding and greater listening effort (Pals et al., 2013, 2020). In our study, the mean SMRT score among adult CI users was 2.06 RPO (SD = 1.34), which is lower than the NH group typically scoring approximately 4 RPO or higher, aligning with prior findings (Henry et al., 2005; Mattingly et al., 2018). These findings support the critical and primary role of bottom-up processing in phonological processing and speech perception outcomes in CI users, even with the audiovisual stimuli provided in this task, which are expected to enhance recall accuracy.

Both TOWRE word and nonword reading efficiency significantly predicted NWRT scores, suggesting that individuals with stronger decoding and lexical access skills also demonstrated better phonological processing on the NWRT. However, age accounted for additional variance in NWRT performance in the TOWRE word model (p = .026), but not in the TOWRE nonword model (p = .148). While TOWRE-2 scores reflect basic reading fluency, word and nonword reading likely engage in different cognitive and linguistic processes. The word-reading domain is closely associated with processing speed, concentration ability, and access to phonological representations in long-term memory and may be less sensitive to the phonological and memory demands involved in the NWRT. In contrast, nonword reading efficiency is related to phonological decoding skills, vocabulary size, sustained attention, and working memory (Tamati et al., 2021). This distinction may help explain why age contributes to NWRT variance in the word reading model but not in the nonword model. Nonword reading may capture much of the age-related decline in phonological processing and working memory, whereas word reading leaves more unexplained variance that can be accounted for by age, as older adults may experience declines in processing speed and concentration (Nittrouer et al., 2016).

In our analysis, vocabulary size (WordFam) influenced phonological processing in CI users. Vocabulary knowledge has previously been identified as a strong predictor of phonological processing, particularly in children up to around age 8 years (Gathercole et al., 1994). This relationship has also been reported to extend into adulthood; nonword repetition in adults has been shown to correlate with the ability to learn novel phonological forms that do not closely resemble familiar words (Atkins & Baddeley, 1998; Gupta, 2003). Limited evidence suggests that this association is stronger in older adults than in younger adults when learning new words in a foreign vocabulary (Service & Craik, 1993). A larger vocabulary is associated with richer and more precise phonological representations, which in turn enhances performance on tasks such as nonword repetition and phonological awareness in both children and adults (Al-Salim et al., 2020; Nittrouer et al., 2014).

Nonverbal reasoning, measured using the Ravens test, also influenced NWRT performance in CI users, suggesting that executive processes contribute to success in this task. While NWRT is considered a measure of phonological short-term memory, the task also engages higher order abilities such as cognitive control, planning, and speech production. For CI users, these demands may be even greater than for their NH peers, since they may rely more heavily on top-down processing due to degraded auditory input. Strong nonverbal reasoning skills may help listeners identify patterns in the degraded signal and fill in gaps, facilitating more accurate repetition of unfamiliar nonwords. The observed associations between NWRT scores and measures of nonverbal reasoning are consistent with previous findings linking broader cognitive functioning to speech recognition outcomes in this population (Gupta, 2003; Moberly & Reed, 2019; Tamati et al., 2021).

Contrary to previous findings, we did not find that working memory capacity, as measured by the visual digit span task, was a significant predictor of NWRT performance. While findings across studies have been mixed and no single cognitive measure consistently predicts speech outcomes, working memory tests (e.g., reading span and digit span) are among the most frequently reported to correlate with performance (Pisoni & Cleary, 2003). Working memory is thought to play a critical role in phonological processing, particularly through the phonological loop component responsible for the temporary storage and manipulation of verbal information. Age-related decline in working memory has been linked to reduced phonological processing and may help explain individual variability in speech perception (Nittrouer et al., 2016). In children, correlations between Digit Span and NWRT performance have been documented (Dillon & Pisoni, 2006; Gathercole, 2006; Gathercole et al., 1994; Pisoni & Cleary, 2003), whereas some previous studies with adult CI users did not find a significant relationship with speech recognition outcomes (Moberly et al., 2017; Moberly & Reed, 2019). The lack of a significant relationship in our study among older CI users may reflect the limited sensitivity of the forward digit span task in capturing the specific aspects of working memory but not short-term memory. Alternatively, other cognitive mechanisms not fully assessed by this measure may be more critical in supporting phonological processing in this population.

To the best of our knowledge, this study includes one of the largest samples of experienced adult CI users compared to NH peers in examining the effects of hearing status and aging on nonword repetition performance across syllable lengths. One limitation of our study is that CI participants completed the NWRT in their best-aided condition (e.g., CI + HA or bilateral CI), which may not fully isolate the performance attributable to the CI alone. Although this approach better reflects real-world listening, it may limit our ability to evaluate how individuals perform with only CIs and the specific influence of spectral degradation. Another limitation of the study is that experienced CI users did not include baseline pre-implantation data to monitor the progress and potential improvement with CI use. Although long-term CI use has been associated with cognitive benefits, the absence of pre-implantation measures limits our ability to attribute postimplantation outcomes to auditory experience only. Finally, while we did not directly measure auditory deprivation, exploratory analyses of demographic factors (age at implantation, duration of deafness, onset of hearing loss) were included to provide context. These variables may serve as proxies for reduced auditory experience but were not strongly related to NWRT performance. These relationships should be further explored through longitudinal studies that include baseline cognitive assessments and more detailed hearing and developmental histories.

Conclusions

This study offers a comprehensive examination of phonological processing in experienced adult CI users with comparisons to NH peers of a similar age range. Our findings demonstrated that both hearing status and age significantly influenced audiovisual NWRT performance. CI users had weaker NWRT performance than their NH peers and showed age-related declines. Among auditory, cognitive, and linguistic predictors, spectrotemporal processing was the strongest predictor of NWRT performance. Cognitive and linguistic factors, such as reading efficiency, lexical access, vocabulary knowledge, and nonverbal reasoning, also contributed meaningfully to individual variability. Additional research that includes baseline cognitive and linguistic data, along with long-term monitoring of CI use, may offer valuable clinical insights for predicting speech performance and identifying areas for targeted training to improve CI users' phonological processing and speech recognition outcomes.

Data Availability Statement

The data sets generated in the current study are available from the corresponding author upon reasonable request.

Funding Statement

This article is published Open Access under a read and publish agreement between Vanderbilt University and the American Speech-Language-Hearing Association. This research was supported by National Institute on Deafness and Other Communication Disorders Grants R01DC019088 (awarded to Aaron C. Moberly) and R21DC019382 (awarded to Terrin N. Tamati), and National Institute on Aging Grant R01AG089200 (awarded to Terrin N. Tamati). Terrin N. Tamati received research funding support from Cochlear America.