Comparing Word Characteristic Effects on Vocabulary of Children with Cochlear Implants

Abstract

Many studies have evaluated overall vocabulary knowledge of children who use cochlear implants, but there has been minimal focus on how word form characteristics affect this knowledge. This study evaluates the effects of neighborhood density and phonotactic probability on the expressive vocabulary of 81 children between five and seven years old (n = 27 cochlear implant users, n = 27 children matched for chronological age, and n = 27 children matched for vocabulary size). Children were asked to name pictures associated with words that have common and rare phonotactic probability and high and sparse neighborhood density. Results indicate that children with cochlear implants, similar to both groups of children with typical hearing, tend to know words with common probability/high density or with rare probability/ sparse density. Patterns of word knowledge for children with cochlear implants mirrored younger children matched for vocabulary size rather than age-matched children with typical hearing.

Children with cochlear implants tend to have lower vocabulary knowledge than their same-age peers (e.g., Lund, 2016; Pimperton & Walker, 2018). Although studies have characterized factors associated with low vocabulary knowledge in children with cochlear implants (e.g., age at implantation, family resources, child cognitive status; Dillon, de Jong, & Pisoni, 2011; Fagan, Pisoni, Horn, & Dillon, 2007; Geers & Sedey, 2011), the content of words known by children with cochlear implants has been only minimally explored (e.g., Nott, Cowan, Brown, & Wigglesworth, 2009; Välimaa, Kunnari, Laukkanen‐Nevala, Lonka, & National Clinical Research Team, 2018). In other words, it is unclear if children with cochlear implants experience a delay in vocabulary knowledge and so know similar words as younger children with normal hearing or if they experience a deficit and know different words than younger children with normal hearing. If, as preliminary work suggests, children with cochlear implants tend to learn different words than children with normal hearing (e.g., Nott et al., 2009), it is critical to understand how hearing loss interacts with word properties to yield a child’s lexicon. Thus, the purpose of this paper is to explore how two word-form characteristics, neighborhood density and phonotactic probability, interact to affect the vocabulary knowledge of children with cochlear implants.

Word Learning and Word Form Characteristics

The process of word learning, even in very young children, is remarkably efficient (Houston-Price, Plunkett, & Harris, 2005). However, most models of learning agree that this process is also complex (e.g., Hirsh-Pasek, Golinkoff, & Hollich, 2000; Leach & Samuel, 2007; Yu & Ballard, 2007). To engage in word learning, a child must first trigger learning by recognizing that a word is novel and needs a referent (Storkel, Armbruster, & Hogan, 2006). Leach and Samuel (2007) then posit that children must experience configuration of a new word to store it in short-term memory. That configuration necessarily involves building a phonological representation (recognizing the individual sounds in the word), a lexical representation (recognizing the full word form), and a semantic representation (recognizing the meaning) of a word. Finally, a child must experience engagement of the new word by integrating it into his or her existing lexicon.

Two word-form characteristics that affect word learning in people with typical hearing include neighborhood density and phonotactic probability. Neighborhood density, related to the lexical representation of a word, refers to how similar an entire word is to other words in one’s lexicon (Hoover, Storkel, & Hogan, 2010). Words with high neighborhood density are those which differ from many other words by only one phoneme. For example, the word bat is similar to cat, bet, back, at, and many others; therefore, it has high neighborhood density. Words with few neighbors, such as orange, have sparse neighborhood density. The second characteristic, phonotactic probability, is related to the phonological representation of a word and refers to the likelihood that a given set of phonemes within a word will occur in that specific order within a language. Bat is a word with common phonotactic probability, as the sounds /b/, /ae/ and /t/ are statistically likely to occur in that order in English. Choice is a word with rare phonotactic probability because the sequence of sounds present in the word is relatively unlikely to occur in English. Neighborhood density and phonotactic probability tend to be highly correlated; that is, high-density words tend to have common sound sequences and words with sparse density tend to have rare sound sequences.

Neighborhood density and phonotactic probability appear to influence different parts of the word learning process. During the early stage of rapid word learning (i.e., triggering), rare-sequence words appear easier for adults to learn than common-sequence words (Storkel et al. 2006). In later stages, however (i.e., engagement), high-density words appear easier for adults to learn than low-density words (Storkel et al. 2006). These findings indicate that rare sound sequences may be beneficial for triggering word learning whereas high-density lexical forms may be beneficial for integration and retention of new words.

In studies of preschool children, children tend to learn words best when those words contain common sound sequences and are high density, as compared to words with rare sequences/low density (Hoover et al. 2010; Storkel & Maekawa, 2005; Storkel, 2001). In examining the effects independently, 4-year-old children more accurately learn rare sound sequences as compared to common sequences when density is held constant. When phonotactic probability is held constant, 4-year-old children learn sparse sequences more accurately than dense sequences at an immediate test point, but, dense sequence retention is higher at testing 1 week later (Storkel & Lee, 2011). In another study that assessed immediate learning and retention of new words, Hoover et al. (2010) found that 5-year-old children experienced a rare-sequence advantage in learning for sparse words both in immediate and retention trials, whereas 3- and 4-year-old children experienced the advantage only in retention. Age effects were not observed for neighborhood density, as the density advantage was consistent across all time points and ages. The authors hypothesized that, unlike adults, children may learn best when the convergence of sparse density and rare sequences trigger learning and when high-density and common sequences support configuration and engagement.

Children with communication disorders are also influenced by word-form characteristics, albeit somewhat differently than children who are typically developing. For example, Stokes (2014) found that the existing expressive vocabulary of late-talking children between the ages of 1 and 2 tended to have a higher percentage of high-density words as compared to children of the same age with larger vocabularies. Preschool children with specific language impairment, who are thought to have particular difficulty with lexical configuration (e.g., Kan & Windsor, 2010), appear to be equally sensitive to density in word-recognition tasks as their typically developing peers, but do not display a particular sensitivity to phonotactic probability (Gray, Pittman, & Weinhold, 2014).

Children with phonological disorders also appear to differ in response to word-form characteristics from children who are typically developing. Storkel, Maekawa, and Hoover (2010) found that preschool children with (n = 24) and without (n = 34) phonological disorders demonstrated similar knowledge of sparse-density/rare-sequence words in a vocabulary probe, but only children without phonological disorders demonstrated high levels of knowledge for high-density/common sequences. Storkel and Hoover (2010) confirmed this finding in a word-learning task: both children with and without phonological disorders learned words with sparse-density and rare sequences most accurately. Children with phonological disorders, however, produced fewer partially correct sound sequences associated with words from sparse neighborhoods than did children without phonological disorders. The authors hypothesized that children with phonological disorders may struggle to create complete lexical and semantic representations of words, even when learning is initially triggered by a sparse-density/rare-sequence combination. Understanding the influence of word-form characteristics on lexical development helps researchers understand not only if vocabulary knowledge differs between impaired and non-impaired populations, but also how it differs.

Vocabulary in Children with Cochlear Implants

To date, the vocabulary delays of children with cochlear implants have been described holistically: children with cochlear implants know fewer words overall than their same-age peers. For example, Lund (2016) in a meta-analysis considered whether the number of words, as measured by naming tasks, known by children with cochlear implants differed from children with typical hearing, and whether that difference magnitude varied according to age or amount of experience listening. Children with cochlear implants evaluated in this meta-analysis had a consistent delay in vocabulary size. Individual studies (e.g., Pimperton & Walker, 2018) confirm this finding: vocabulary can be an area of weakness for children with cochlear implants as compared to children with typical hearing. This weakness is of concern for professionals working to improve language outcomes for children with hearing loss because vocabulary is a building block for the development of other linguistic skills (e.g., grammar).

It remains unclear how word form characteristics (e.g., neighborhood density and phonotactic probability) affect the vocabulary knowledge of children with cochlear implants. Those studies that address neighborhood density and phonotactic probability primarily consider the effects of those characteristics on speech perception. For example, Kirk, Pisoni, and Osberger (1995) determined that children with cochlear implants recognized words with sparse neighborhood density more easily during speech perception tasks than high-density words. This result has been replicated (e.g., Dirks, Takayanagi, & Moshfegh, 2001; Eisenberg, Martinez, Holowecky, & Pogorelsky, 2002) and is perhaps intuitive: it is easier to identify a word that does not sound like other words than to identify many words that sound very similar.

Two restrospective studies to date have evaluated how phonotactic probability and neighborhood density relate to the holistic vocabulary knowledge for children with cochlear implants. Han, Storkel, Lee, and Yoshinaga-Itano (2015) conducted a retrospective analysis of vocabulary measures administered to 14 children with cochlear implants at the age of 2, 3, and 6 (reported via vocabulary checklist and norm-referenced measures). Across each of those ages, children with cochlear implants were more likely to know high-density words than sparse-density words. However, phonotactic probability did not appear to affect the vocabulary knowledge children with cochlear implants, unlike children and adults with typical hearing. This study is the first to suggest that degraded auditory input perhaps limits triggering of word learning (affected by phonotactic probability) but not storage (affected by neighborhood density). Guo, McGregor, and Spencer (2015) used archival data (n = 36 children 12 months post-cochlear implantation with a mean age of 27 months) to determine that the vocabulary size of children with bilateral cochlear implants was positively correlated with the phonotactic probability of words and negatively correlated with the average neighborhood density of words. However, children with unilateral cochlear implants in the Guo and colleagues study did not appear to have this same sensitivity to word-form characteristics. Neither study was able to demonstrate the separate effects of neighborhood density or phonotactic probability on vocabulary knowledge of children with cochlear implants. The available evidence does suggest that it is possible, and perhaps even likely, that the degraded auditory signal experienced by children with cochlear implants affects the characteristics of words they learn and retain.

The purpose of the present study was to evaluate the knowledge of children with cochlear implants with respect to neighborhood density and phonotactic probability. In the same way, Storkel et al. (2010) used a vocabulary probe to differentiate the effects of word-form characteristics on the knowledge of children with phonological delays, this study implemented a vocabulary probe to systematically assess neighborhood density and phonotactic probability of children with and without cochlear implants. Initially, the researchers sought to compare the validity of vocabulary assessment via probe to vocabulary assessment via a published, norm-referenced measure. The first research question asked (a) Do children with cochlear implants know fewer words than children with typical hearing matched for age, regardless of whether vocabulary knowledge is measured by a norm-referenced assessment or by a vocabulary probe that systematically varies the phonotactic probability and neighborhood density of target words?

Using only the probe, this study then sought to evaluate independently the effects of word form characteristics on vocabulary knowledge of children with cochlear implants: (b) Do children with cochlear implants know proportionally fewer rare-sequence words than children with typical hearing matched for age or for vocabulary size when density is controlled? (c) Do children with cochlear implants know proportionally more sparse-density words than children with typical hearing matched for age or for vocabulary size when phonotactic probability is controlled?

Finally, given the findings of the Guo et al. (2015) differentiating the knowledge of unilateral versus bilateral cochlear implant users, this study addressed the questions: (d) Does the effect of phonotactic probability and neighborhood density vary based on use of bilateral versus unilateral cochlear implants? and (e) Does vocabulary size as measured by a norm-referenced measure correlate with knowledge of words based on phonotactic probability and neighborhood density?

Method

This study was approved by the Texas Christian University Institutional Review Board.

Participants

Eighty-one children participated in this study: 27 children with cochlear implants (CI group), 27 children matched for chronological age to children in the cochlear implant group (CA group), and 27 children matched for vocabulary size (VM group) as measured by the raw score on the Expressive One Word Picture Vocabulary Test—Fourth Edition (EOWPVT-4; Martin & Brownell, 2010a) within 5 points of a child in the cochlear implant group.

Inclusionary criteria for the CI group in this study included use of at least one cochlear implant device and speech perception skills at the level of Consistent Word Identification according to the CID Early Speech Perception Test (Moog & Geers, 2012). Children with Consistent Word Identification are able to consistently point to similar-sounding consonant–vowel–consonant words presented through audition alone from a set of 12 options. Two children had participated in the cochlear implant group experimental protocol but are not included in this study because their speech perception skills, despite many years of cochlear implant use, remained only at the level of Pattern Perception (correctly identified words varying in syllable length but not words that differed at the phoneme level).

All children participated in a descriptive test battery that included the EOWPVT-4, the Receptive One Word Picture Vocabulary Test—Fourth Edition (ROWPVT-4; Martin & Brownell, 2010b), the Primary Test of Nonverbal Intelligence (Ehrler & McGhee, 2008), the Test of Early Language Development (Hresko, Reid, & Hammill, 1999), the CID Early Speech Perception Test and the Arizona Articulation Proficiency Scale—3 (Fudala, 2000). Children in the groups with typical hearing (AM, VM) completed a hearing screening at 25 dBHL across 500, 1,000, and 200 Hz frequencies. Group characteristics and performance on descriptive measures are listed in Table 1.

Table 1.

Group characteristics

Group	Age	Race	Ethnicity	Years of maternal education	PTONI standard score	TELD standard score	AAPS standard score	ROWPVT standard score	EOWPVT standard score	EOWPVT raw score
CI (n = 27)	73.81 (10.24)	22 White	7 Hispanic	16.44 (2.12)	102.85 (15.20)	86.30 (15.37)	86.74 (9.44)	95.63 (11.60)	94.14 (14.80)	69.35 (13.80)
		4 Black	20 Nonhispanic
		1 Asian
AM (n = 27)	73.26 (9.28)	24 White	5 Hispanic	17.22 (1.81)	105.59 (18.39)	110.92 (7.58)	96.33 (6.39)	121.19 (9.69)	115.74 (10.65)	95.33 (10.38)
		2 Black	22 Nonhispanic
		1 Asian
VM (n = 27)	58.70 (10.81)	20 White	8 Hispanic	17.48 (1.28)	102.63 (16.71)	109.77 (13.87)	96.26 (9.51)	110.15 (10.21)	106.70 (9.99)	66.41 (11.38)
		3 Black	19 Nonhispanic
		4 Asian

Note. PTONI = Primary Test of Nonverbal Intelligence (Ehrler & McGhee, 2008), TELD = Test of Early Language Development (Hresko et al. 1999), AAPS = Arizona Articulation Proficiency Scale (Fudala, 2000), ROWPVT = Receptive One Word Picture Vocabulary Test (Martin & Brownell, 2010b), EOWPVT = Expressive One Word Picture Vocabulary Test (Martin & Brownell, 2010a), CI = cochlear implant group, AM = age-matched group, VM = vocabulary matched group

Children in the cochlear implant group all reported using spoken English as a primary mode of communication. Each of the children’s parents also reported using spoken English as a primary mode of communication (i.e., no parent only used sign language). Several parents reported that their children had been introduced to sign language in infancy and during preschool, but all parents also reported that currently they only communicate with their children using spoken language. Additional diagnoses of children were also reported by parents, and included ADHD, hypothyroidism, sensory processing disorder, and a seizure disorder. No child in this group had a diagnosis known to be associated with delayed linguistic or overall nonverbal cognitive development (e.g., Autism Spectrum Disorder, Down syndrome). Seven children were implanted with an Advanced Bionics device, 16 wore a Cochlear Americas brand device, and 6 wore a Med-El brand of device. Seventeen children had bilateral cochlear implants and ten wore only unilateral cochlear implants. Of those children who wore unilateral cochlear implants, only two wore a hearing aid in the un-implanted ear. Hearing loss was not identified for 5 of the 27 children until after those children were one year old (range of identification: birth—30 months, average age at identification 6.5 months); thus, it is possible that these children experienced a progressive hearing loss. Of those five children, two were identified after age 2. Children ranged in age from 60 months to 90 months, and amount of time with access to the full range of speech sounds ranged from 30 months to 78 months.

Task Development

Creation of the vocabulary probe was informed by Storkel et al. (2010). From the Storkel (2013) corpus of 804 consonant–vowel–consonant words (all real words found in the child corpus (Storkel & Hoover, 2010), 857 words were selected from the highest quartile and the lowest quartile of neighborhood density (according to child corpus statistics). From those lists, words that fell within the highest quartile of phonotactic probability (common probability words; n = 271 words of high density and 23 words of low density as measured by the positional segment sum of the word based on a child corpus of word knowledge) and the lowest quartile (rare probability words; n = 22 words of high density and 275 words of low density) were selected. Additionally, for those lists, word frequency (SUBTLEXus; Brysbaert & New, 2009) and age-of-acquisition rating (Kuperman, Stadthagen-Gonzalez, & Brysbaert, 2012) were calculated. Highly picture-able words were then selected from each group of words (high density/common probability, high density/ rare probability, sparse density/common probability and sparse density/rare probability and images representing each word were created.

Only words with an estimated age of acquisition between age three and age ten were selected. Images for 100 words selected were shown to 15 adults with no history of language disorder per self-report, and these adults were asked to label the pictures. Only those pictures that elicited the target word from all adults were included in the final probe. The final experimental probe included 20 words with high density and common probability, 10 words with high density and rare probability, 20 words with sparse density and rare probability, and 10 words with sparse density and common probability. Age of acquisition was compared across lists and found to not be significantly different (p = .75). Pictures were added to a Microsoft Powerpoint program and randomly distributed (i.e., the final order of the pictures in the probe was randomized). Table 2 displays characteristics of word lists included in the final probe.

Table 2.

Probe list characteristics

List	Number of neighbors	Phonotactic probability	Age of acquisition	Frequency
High density/common probability	20.96 (3.70)	.23 (.03)	5.59 (1.77)	43.53 (95.60)
High density/rare probability	17.60 (2.12)	.13 (.01)	5.39 (1.09)	51.21 (77.57)
Sparse density/rare probability	6.73 (1.43)	.11 (.02)	5.79 (1.66)	32.31 (65.57)
Sparse density/common probability	7.60 (2.31)	.20 (.01)	5.20 (1.09)	42.68 (74.23)

Note. Number of neighbors and phonotactic probability (positional segment sum) derived from Storkel (2013) in comparison to child corpus; age of acquisition represented in years from Kuperman et al. (2012); frequency from Brysbaert & New (2009)

Procedures

In this study, all children completed the standardized/ descriptive measures and the experimental naming task in one or two sessions (if two, the sessions were scheduled within a week of each other). Children sat in front of a 14-inch display computer screen, next to the examiner. The examiner, who was the author or a trained research assistant, showed the series of pictures to the participant, who was asked to name each picture. If the child produced a response that was close to the target (e.g., pond for lake), the child was re-prompted with “Yes, but can you think of another word for this picture?” Responses were phonetically transcribed and marked as correct or incorrect and video-recorded. Children were not penalized for articulation errors that were consistent with errors made during other testing (e.g., wabbit for rabbit).

A second scorer watched a video and audio recording of responses to each of the pictures and separately scored those videos. These scores were compared with the original examiner’s score to compute point-by-point reliability. Simultaneously, the video reviewer observed the examiner’s behavior and marked whether the examiner appropriately administered the prompts to the child (used the correct, scripted prompts (e.g., what is this?) and correctly re-prompted children). Examiners administered the probe with greater than 99% fidelity. Point-by-point agreement for correct/incorrect responses was 100% for this task.

Analysis

To answer the first research question, t-tests were used to compare group responses on the norm-referenced and on the experimental expressive vocabulary measure. Additionally, a correlation analysis was calculated to compare results across the two measures.

To answer the second, third, and fourth research questions, repeated measures analysis of variance calculations were used. Within these analyses, the dependent variable was percent words correctly named, and within-subjects variables included density status of words (high versus sparse), and phonotactic probability (common versus rare). Between-subjects variables included group membership according to the research question (e.g., CI, AM or VM group; unilateral or bilateral implant user). Main effects and interaction effects were calculated using Tukey tests, and Bonferroni corrections to the expected p value to denote significant differences were applied as appropriate. Within the final research question, which applied only to children within the CI group, time spent listening and age of implantation were considered as covariates.

Results

This study sought to evaluate the word knowledge of children with cochlear implants as compared to peers with typical hearing with respect to the word characteristics of neighborhood density and phonotactic probability. The first research question addressed whether children with cochlear implants knew fewer vocabulary words than children with typical hearing matched for age across both the norm-referenced measure and the probe measure. To answer this question, the expressive norm-referenced measure used in this study, the Expressive One Word Picture Vocabulary Test, was used to derive standard scores for each participant in the cochlear implant group and each participant in the age-matched group. These standard scores were entered into a t-test as dependent variables. Results revealed a significant group difference t(52) = 8.79, p < .001, d = 2.39, indicating children with cochlear implants, consistent with other studies, scored lower than children with typical hearing matched for chronological age on an omnibus, norm-referenced measure of vocabulary knowledge. See Figure 1.

Figure 1.

Expressive one word picture vocabulary test standard scores by group.

To then assess the validity of the current vocabulary probe constructed to measure vocabulary knowledge across neighborhood densities and phonotactic probabilities, the number of total words correct for each participant on the experimental task was calculated across all categories of words (high density and common probability, high density and rare probability, sparse density and rare probability, sparse density and common probability) as the dependent variable in a second t-test. Again there was significant difference between children with cochlear implants and children with typical hearing matched for age (t(52) = 7.29, p < .001, d = 1.99). See Figure 2. Further support for the probe measure’s validity is demonstrated by the significant correlation between EOWPVT score and the experimental task score (r(52) = .71, p < .001).

Figure 2.

Total correct on vocabulary probe by group.

To answer the second and third research questions, proportion of words correctly identified on the probe was entered as the dependent variable in a repeated measures analysis of variance, with phonotactic probability (common versus rare) and neighborhood density (high versus sparse) entered as within-subjects independent variables and group (AM, VM or CI) as a between-subjects variable. This analysis yielded a main effect of density (F(1,78) = 7.508, p = .008) and of group (F(2, 78) = 36.93; p < .001) but not probability (F(1, 78) = 1.47, p = .23). Interaction effects were found for density and group (F(2, 78) = 4.06, p = .02) and for density and probability (F(1, 78) = 152.09, p < .001). Follow-up Tukey tests were evaluated against a Bonferroni-corrected p value for multiple comparisons (p < .016) and indicated that, relative to overall vocabulary knowledge, the AM group (p < .001) differing significantly from the CI and VM group, which did not significantly differ from each other (p = .901). Relative to density, children overall, regardless of group membership, tended to know more sparse-density words than high-density words (p < .001). The interaction between density and probability (evaluated against the multiple-comparison p value corrected to p < .005) reflects that children in each group were more likely to know words where probability and density were matched (e.g., high density and common probability) than when they were mismatched (e.g., high density and rare probability; p < .001). Finally, the interaction effect of density and group indicated that children in the AM group tended to know more sparse-density words than high-density words, whereas children in the CI and VM group tended to know similar numbers of high and sparse-density words (p < .001). See Figure 3.

Figure 3.

Words known by each group according to neighborhood density and phonotactic probability.

The final research questions addressed whether the effect of phonotactic probability and neighborhood density varied based on use of bilateral versus unilateral cochlear implants. This question was motivated by prior works indicating that unilateral use of a cochlear implant may affect word learning differently than bilateral use (Guo et al., 2015). Only the cochlear implant group results were entered into another repeated measures analysis of variance, again with density and phonotactic probability as within-subjects factors and cochlear implant status (unilateral versus bilateral) as a between-subjects factor. Age at cochlear implant activation and experience listening with a cochlear implant are factors commonly associated with linguistic outcomes for children with cochlear implants (Connor, Craig, Raudenbush, Heavner, & Zwolan, 2006). Both were evaluated as covariates, and both age at implant activation and time listening with a cochlear implant were significantly correlated with the dependent variable (r(26) =−.46, p = .016; r(26) = .52, p = .005 respectively). Both were added as covariates in analysis. There was no main effect of cochlear implant status (F(1,23) = .11, p = .742) but the interaction effect of density and probability remained (F (1, 23) = 13.86, p = .001). See Figure 4.

Figure 4.

Words known by unilateral and bilateral cochlear implant users according to phonotactic probability and neighborhood density.

To compare these findings with those of Guo et al. (2015), the effect of general vocabulary knowledge as measured by the EOWPVT on word knowledge across all density and probability categories was evaluated via correlation analyses that evaluated children with unilateral cochlear implants and bilateral cochlear implants separately. For children with unilateral cochlear implants, there were no significant correlations between EOWPVT standard score and high density/common probability words (r(9) = .47, p = .173), high-density/rare probability words (r(9) = .28, p = .439), sparse-density/rare probability words (r(9) = .61, p = .062) or sparse-density/common probability words (r(9) = .59, p = .080). It is possible, however, that this analysis was somewhat underpowered: only 10 children’s results could be included in this correlation analysis. For children with bilateral cochlear implants, there was a significant correlation between EOWPVT standard score and sparse-density/rare probability words (r(9) = .62, p = .008) and high-density/ common probability words (r(9) = .54, p = .027), but not high-density/rare probability words (r(9) = .39, p = .125) or sparse-density/common probability words (r(9) = .09, p = .724).

Discussion

This study sought to evaluate the effects of word form characteristics on the expressive vocabulary knowledge of children with cochlear implants as compared to children with typical hearing matched for age and matched for vocabulary size. analyses yielded findings related to general vocabulary knowledge of children with cochlear implants, the effects of neighborhood density and phonotactic probability on vocabulary knowledge of children with cochlear implants, and the effects of unilateral versus bilateral implantation on vocabulary knowledge.

Overall Vocabulary Knowledge

Across all measures of vocabulary knowledge, children with cochlear implants in this study scored lower than children in the age-matched group with typical hearing. The effect size difference was large (d = 2.39) between the standard scores for groups on the EOWPVT-4, a norm-referenced measure, which is consistent with previous studies of children with cochlear implants (e.g., Pimperton & Walker, 2018). Notably, the average standard score for children with cochlear implants was 94.14 (SD = 14.80), which falls well within the “range of normal” performance for children according to test norms alone. This finding highlights the need for comparison groups matched for variables beyond age: children in this study were matched geographically and relative to nonverbal cognitive scores in addition to age. Others have called for more nuanced analysis of language outcomes for children with hearing loss beyond use of norm-referenced scores (Lund, 2016; Werfel & Douglas, 2017). This study provides an additional instance of children with cochlear implants scoring within the “range of normal” but still clearly performing differently than their age-matched peers.

The statistically significant main effect of group on the experimental vocabulary measure also indicated that children with cochlear implants have lower vocabulary knowledge than children with typical hearing matched for age. That difference between these two groups (CI and AM) was consistent regardless of word density or phonotactic probability. However, the scores of children with cochlear implants on the experimental measure were highly similar to the scores of children matched for vocabulary size via the EOWPVT-4 raw score. There was also a strong, significant correlation between EOWPVT-4 scores and the experimental task scores. These findings suggest that the experimental measure was a valid measure of overall vocabulary knowledge: children matched for vocabulary size on an established measure of expressive vocabulary performed similarly to their matched group on the experimental task. As indicated by Storkel et al. (2010), a vocabulary probe with targets that are intentionally divided by neighborhood density and phonotactic probability still captures group differences in overall vocabulary knowledge.

Effects of Word form Characteristics

Despite an overall difference in vocabulary knowledge for children in the AM group as compared to the VM and CI groups, there was a significant interaction effect between neighborhood density and phonotactic probability across all groups, regardless of group membership. Children were more likely to have words in their vocabulary that were matched for neighborhood density and phonotactic probability than words that were unmatched. In other words, children knew more targets that had both high density and common probability or sparse density and rare probability than words that were high density and rare probability or sparse density and common probability.

This finding is very consistent with the work of Hoover et al. (2010), who hypothesized that matched neighborhood density and phonotactic probability created ideal word-learning conditions for children. Although children with cochlear implants knew fewer words across each category (matched and unmatched) as compared to children with typical hearing who were the same age, the fact that their word knowledge patterns were similar relative to these word form characteristics is encouraging. It appears the same types of words (i.e., words matched for neighborhood density and phonotactic probability) trigger learning and retention for children with and without cochlear implants. These results for children with cochlear implants appear different from those for children with phonological disorders (Storkel & Hoover, 2010; Storkel et al. 2010). Children with phonological disorders appear to learn words with sparse density and rare probability as well as children without phonological disorders, but they do not demonstrate similar levels of high-density and common probability word knowledge. Storkel and Hoover (2010) hypothesized that children with phonological disorders trigger word learning similarly to children without phonological disorders but then struggle to complete lexical and semantic representations of words. The present results indicate that children with cochlear implants may not struggle specifically with configuration of new words once learning is triggered (as might be indicated by a discrepancy in word knowledge for sparse-density/rare probability words versus high-density/common probability words).

This study did not find an independent effect of phonotactic probability overall or within any group. Other studies that have explored the effect of phonotactic probability separate from the effect of neighborhood density on word-learning yield different results (e.g., Gray et al. 2014; Storkel & Lee, 2011; Storkel et al. 2006). However, this discrepancy may be the result of methodological differences: the present study evaluated existing word knowledge whereas other studies finding a particular effect of phonotactic probability assessed the effects of word form characteristics on active word learning. Storkel et al. (2010) found similar results to this study using a vocabulary probe: children who were typically developing showed higher knowledge of words with matched neighborhood density and phonotactic probability than words with unmatched characteristics. Han et al. (2015) found similar results for children with cochlear implants: phonotactic probability independently did not appear to affect word knowledge for children between the approximate ages of 2 and 6 years.

Relative to neighborhood density, there was a significant interaction effect between group and density. This effect was driven by the tendency of children in the age-matched group to know more sparse-density words than high-density words, whereas children in the cochlear implant and vocabulary matched groups knew roughly the same numbers of high and sparse-density words. Again, this finding is not supported by studies of word-learning (e.g., Storkel & Lee, 2011), but the findings for children in the vocabulary matched and the cochlear implant group are consistent with work using a vocabulary probe (Storkel et al. 2010). The existence of a group difference relative to neighborhood density again speaks to the delay in vocabulary knowledge experienced by children with cochlear implants: their profiles of word knowledge appear more like those of younger children.

Effects of Unilateral vs. Bilateral Cochlear Implantation

The ten children in this study who wore only one cochlear implant did not appear to have different amounts of word knowledge relative to word form characteristics as compared to the other seventeen children who wore two cochlear implant devices. This pattern of performance was consistent even when age at first implantation and time listening with a cochlear implant were accounted for as covariates. However, the expressive vocabulary knowledge of children in the bilateral group, as measured by the EOWPVT, significantly correlated with vocabulary knowledge across matched neighborhood density and phonotactic probability categories, whereas the EOWPVT standard score of children in the unilateral group did not correlate with any outcomes on the experimental vocabulary measure.

These results are consistent with the findings of Guo et al. (2015), who found that there was not an effect of unilateral/bilateral cochlear implantation on total words known, but that the correlation between vocabulary size and words with differing word form characteristics was only significant for children with bilateral cochlear implants. Guo and colleagues were not able, however, to partial out how phonotactic probability and neighborhood density might interact to support word knowledge of children with cochlear implants. The present study adds to their findings by demonstrating that (a) there is a specific relation between general expressive vocabulary knowledge and matched neighborhood density/phonotactic probability word knowledge for children with bilateral cochlear implants but not for children with unilateral cochlear implants and (b) that the unilateral versus bilateral difference found by Guo et al. (2015) in children 1 year post-implant appears to be maintained many years later. It is possible that children with unilateral cochlear implants, who can be at a disadvantage compared to children with bilateral implants for speech perception (e.g., Grieco-Calub & Litovsky, 2012), are not as sensitive to word-form characteristics as children with bilateral cochlear implants. However, it also appears that the lack of sensitivity does not necessarily make children with unilateral cochlear implants unlikely to learn as many total words as children with bilateral cochlear implants.

Clinical Implications

Overall, these data indicate that children with cochlear implants know fewer words than children with typical hearing of the same age, but the same types of matched word form characteristics (e.g., high density and common probability) converge to support word learning in children with and without cochlear implants. Notably, children with cochlear implants appear to perform similarly to younger children (more than a year younger on average) matched broadly for vocabulary size. A possible conclusion is that word form characteristics do not cause children with cochlear implants to retain different words from all children with typical hearing; children with cochlear implants appear to experience a delay in vocabulary knowledge as compared to age-matched peers.

Although the finding that children with cochlear implants experience a delay rather than difference in vocabulary knowledge relative to word form characteristics is positive, differences between children with cochlear implants and children matched for age are still cause for concern. Sixteen of the children in the cochlear implant group had omnibus language scores that fell less than one standard deviation below the test mean; therefore, these students were likely not eligible for additional language support services at school. Even when performance on other measures of language is taken into account, such as articulation and vocabulary size, 14 of the 16 children still would not qualify for services (two children scored below 85 on the measure of articulation skills). From the perspective of “qualification” for services, these 16 children are treated as children with typical hearing of the same age, and all of them participated in mainstream classroom settings. These children are therefore being taught in classrooms with their peers and expected to learn the same material as peers with typical hearing, but they are required to do so with lower vocabulary knowledge.

A potential problem with the use of norm-referenced measures as a means of determining who does and does not receive clinical services lies in the fact that norm-referenced measures can only match children according to age. Other variables, such as maternal education level and geographic area contribute substantially to vocabulary outcomes for children (e.g., Dollaghan, Campbell, Paradise, Feldman, Janosky et al., 1999; Qi, Kaiser, Milan, & Hancock, 2006). For example, nationally representative age-norms alone might not be consistent with the norms for a specific geographic region. Within the present data set, children were recruited from the same geographic community and had similar socioeconomic backgrounds. Children in the groups with normal hearing had a mean omnibus language score of approximately 110 and the age-matched group achieved mean vocabulary scores of 115–120 (whereas the national age-norms would predict the average should be around 100). This finding is consistent with other studies comparing children with cochlear implants to a group of children with normal hearing rather than using a normative comparison alone (Lund, 2016; Werfel & Douglas, 2017). The inherent expectation of a mainstream classroom using norm-referenced measure cut-offs to determine support service eligibility is that children with a language standard score of 85 have the necessary skills to progress through the academic curriculum. Although children with normal hearing who scored similarly to children with hearing loss within this community would not receive additional support services (i.e., would not be eligible for speech-language pathology support services), educators may want to consider that children with cochlear implants experience cognitive, social and linguistic developmental trajectories that differ somewhat from children with normal hearing (e.g., Houston et al. 2012). Therefore, children with cochlear implants, even those who achieve standard scores within the “range of normal” may continue to need support to fully access the academic curriculum.

Differences in vocabulary knowledge for children with cochlear implants may have later effects on academic development. Vocabulary knowledge, as a part of oral language knowledge, is an important predictor of later reading outcomes (Kyle & Harris, 2011; Senechal, Ouellette, & Rodney, 2006; Storch & Whitehurst, 2002). It may also contribute to the development of other early literacy skills. For example, the Lexical Restructuring Hypothesis (Garlock, Walley, & Metsala, 2001) predicts that children who develop dense neighborhoods of vocabulary must begin to attend to the phonological structure of words. This attention to word structure separate from meaning is often referred to as phonological awareness (Schuele & Boudreau, 2008) and is a key component of early literacy development. If children with cochlear implants know fewer dense words than children with normal hearing, we might predict that children with normal hearing would develop phonological awareness more readily than children with cochlear implants. This prediction is supported by data on phonological awareness skills in children with cochlear implants (e.g., Nittrouer, Sansom, Low, Rice, & Caldwell-Tarr, 2014).

Beyond the development of early literacy skills, vocabulary knowledge is also likely to impact subject areas such as mathematics, social studies and science, which each involve “tiers” of vocabulary knowledge (Beck, McKeown, & Kucan, 2002). Teachers in mainstream classrooms are likely to teach words that are subject-specific (e.g., settlement in social studies), but may also assume that more basic words, or Tier 1 words, are common knowledge. If children with cochlear implants know fewer words than children with typical hearing, they may need more vocabulary support than their peers, and, therefore, fall behind academically. Educational professionals with knowledge of language development and hearing loss, such as speech-language pathologists or teachers of the deaf, may be well-positioned to advocate for support services for children who score in the “range of normal” on omnibus tests of language, or rather, to encourage assessing children’s language with more than norm-referenced tests (e.g., Werfel & Douglas, 2017).

Future Directions and Limitations

Limitations of this study provide opportunities for future research on this topic. The first limitation lies in form of assessment used by this study: vocabulary knowledge is a complex construct that cannot be fully assessed by a naming task. A possible reason that findings from this study (e.g., that phonotactic probability did not affect performance of children with or without cochlear implants) do not fit with findings from other studies is related to the probe methodology used to assess vocabulary. Many other studies (e.g., Gray et al., 2014; Storkel & Lee, 2011; Storkel et al. 2006) have assessed the effects of word form characteristics have on word learning (a process) rather than words known and retained (words assessed in the probe). Future works can assess the effects of word form characteristics on other measures of the vocabulary knowledge construct (e.g., the word learning process.

The sample of children used in this study also presents a possible limitation to the generalization of these study results: children with cochlear implants were required to have good speech perception outcomes. Not all children with cochlear implants achieve good audibility with cochlear implants (e.g., Davidson, Geers, & Nicholas, 2014), so the data collected as part of this study cannot be generalized to children who have poor audibility with a cochlear implant, or to children who have additional disabilities related to language or cognitive development. Future work could extend this work to assess the effect of word form characteristics on children with cochlear implants who have poor audibility or children with additional disabilities. Similarly, children in this study were primarily spoken language users. Future studies may consider the effect of word form characteristics on word learning of children who are bilingual, using both sign language and spoken language as modes of communication.

Additionally, the conclusions that can be drawn about the effects of bilateral versus unilateral cochlear implantation may be limited. Only 10 of 27 children in this sample used a unilateral cochlear implant. Although the correlations between word form characteristics and vocabulary knowledge were not significant for children with cochlear implants, some of those correlations approached significance (p value between .06 and .08). A study with more children with unilateral cochlear implants may be better able to describe how unilateral versus bilateral implantation affects sensitivity to word form characteristics in learning.

Further study of this topic should also explore how knowledge of words with particular form characteristics affects other academic outcomes in children with cochlear implants. For example, if having low knowledge of high-density words affects children’s literacy outcomes in the classroom, high-density word forms could be a target of intervention for professionals who work with children with hearing loss. Similarly, findings related to matched versus unmatched word form characteristics might help educators to predict which words will be more easily learned in the classroom and which words may require additional explicit instruction. Further, the word form characteristics neighborhood density and phonotactic probability should be explored in relation to semantic features of words that children with cochlear implants learn.

Conclusion

This study represents a step towards understanding not only how many words children with cochlear implants know but also which words they know. Children with cochlear implants in this study appeared to experience a delay in vocabulary knowledge relative to the word form characteristics of neighborhood density and phonotactic probability, in that they had similar performance to younger children with typical hearing rather than different performances. Further, the results of this study support the idea that children with cochlear implants may still experience deficits in their vocabulary knowledge compared to same-aged peers even when their norm-referenced test scores fall in the “range of normal.” Awareness of how word characteristics affect the vocabulary knowledge of children with cochlear implants and future exploration of how word characteristics affect the learning process could positively contribute toward improving outcomes for children with cochlear implants who are learning spoken language.

Funding

National Institutes of Health—National Institute on Deafness and Other Communication Disorders [5R03DC015078 to E. L.].

Conflict of interest

No conflicts of interest were reported.