Probing Phonological Processing Differences in Nonword Repetition for Children with Separate or Co-occurring Dyslexia and Developmental Language Disorder

Abstract

We assessed nonword repetition (NWR) skills in 7–9 year-old children with dyslexia (dyslexia-only), developmental language disorder (DLD-only), co-occurring DLD+dyslexia, and typical development (TD) with a norm-referenced and an experimental task. The experimental task manipulated phonemic variability (dissimilarity among consonant phonemes within the nonword) and presentation modality (audio-only versus audiovisual) to probe potential phonological processing differences among the groups. Across tasks, the dyslexia-only and DLD-only groups performed similarly to each other and intermediately to the TD and DLD+dyslexia groups. In the experimental task, nonwords with low phonemic variability were produced less accurately in both modalities, and audiovisual presentation facilitated accurate repetition of low phonemic variability nonwords. A lack of a group interaction with phonemic variability or presentation modality suggests similarities, despite group differences, in how underlying phonological representations influence task performance. Overall, results suggest that poor NWR is associated with both dyslexia and DLD, and that co-occurrence compounds this difficulty.

Nonword repetition (NWR) tasks are common in research focused on language acquisition and developmental reading and language disorders. NWR is broadly conceptualized as a measure of phonological processing. However, such processing may reflect a variety of factors, including the quality of phonological representations and memory processes, which are also influenced by properties of the nonword stimuli (Dollaghan & Biber, 1993; Gathercole, Willis, Baddeley, & Emslie, 1994). NWR tasks are correlated with other phonological processing tasks, including measures of phonological awareness (Barbosa, Miranda, Santos, & Bueno, 2009; Bowey, 1997; Garlock, Walley, & Metsala, 2001), which are known to predict reading ability (Hogan, Catts, & Little, 2005). NWR has importance beyond the assessment of phonological processing as performance also predicts current and future vocabulary size (Gathercole & Baddeley, 1989; Gathercole, 2006; Hoff, Core, & Bridges, 2008) as well as word-learning abilities (Adlof & Patten, 2017; Gathercole, Hitch, Service, & Martin, 1997; Jackson, Leitao, Claessen, & Boyes, 2019; Montgomery, Magimairaj, & Finney, 2010).

Numerous studies have shown that children with language and reading impairments have difficulty with NWR compared to their typically developing peers (cf. Graf Estes, Evans, & Else-Quest, 2007; Melby-Lervag & Lervag, 2012). Based on these findings, NWR has been implicated as an important indicator of language-learning abilities as well as a marker of risk for language or reading impairment (Dollaghan & Campbell, 1998; Redmond, Thompson, & Goldstein, 2011; Saiegh-Haddad & Ghawi-Dakwar, 2017). However, previous research has been unable to distinguish if difficulty on NWR tasks for children with language or reading disorders is due to impairment in the same underlying processing mechanism. The current study was designed to begin to probe this question.

NWR in Children with Dyslexia or Developmental Language Disorder

Meta-analyses of NWR in children with reading or language impairments have found poorer group performance for children with dyslexia (Melby-Lervag & Lervag, 2012) or developmental language disorder (DLD; Graf-Estes, et al., 2007) relative to typically developing (TD) peers. However, few studies included in the meta-analyses accounted for the frequent co-occurrence of dyslexia and DLD (Catts, Adlof, Hogan, & Weismer, 2005), which complicates the interpretation of these results. It remains unclear whether NWR difficulty is unique to one of these disorders. Furthermore, it is unclear if the underlying source of NWR impairment is the same or different for dyslexia and DLD.

Dyslexia is a reading disorder characterized by difficulties with accurate or fluent word decoding and spelling (American Psychiatric Association, 2019; Lyon, Shaywitz, & Shaywitz, 2003) that are not explained by inadequate instruction. These word-level reading difficulties may lead to secondary problems with reading comprehension (Gough & Tunmer, 1986; Hoover & Gough, 1990; International Dyslexia Association, 2012). Deficits in the phonological component of language are considered core features of the dyslexia profile. These may include under-specified phonological representations (Elbro & Jensen, 2005; Farquharson, Centanni, Franzluebber & Hogan, 2014), poorer phonemic awareness (Torgesen, Wagner, & Rashotte, 1994), and poor phonological memory (Menghini, Finzi, Carlesimo, & Vicari, 2011; Schuchardt, Bockmann, Bornemann, & Maehler, 2013). These phonological deficits are thought to impede the acquisition of letter-sound correspondences that are necessary for acquiring decoding skills.

DLD is defined as a difficulty in the ability to comprehend and produce spoken language despite adequate environmental stimulation in the absence of intellectual disability, acquired head injury, or other biomedical diagnoses that are known to affect language development (Bishop, Snowling, Thompson, Greenhalgh, & CATALISE-2, 2016; 2017). Importantly, much of the research we cite in this paper related to DLD involved children with specific language impairment (SLI). The term DLD was recently proposed as a replacement for the disorder known SLI. SLI and DLD are conceptually similar, but most studies of children with SLI have excluded children with low nonverbal IQ scores (e.g., more than 1 SD below the mean; Gallinat & Spaulding, 2014; Tomblin, Records, Buckwalter, Zhang, Smith, & O’Brien, 1997), whereas the consensus statement on DLD explicitly includes children with low nonverbal intelligence who are not intellectually disabled (Bishop et al., 2017). Thus, SLI can be considered a subgroup of DLD (e.g., Norbury, et al., 2016).

Children with DLD may exhibit deficits in multiple domains of oral language, including phonology, semantics, morphology and syntax, pragmatics, and discourse (Bishop, et al., 2017; Tomblin, et al., 1997). There is no domain in which all children with DLD are required to have difficulty, but prior research has identified grammatical difficulties, especially tense and agreement marking in children who speak mainstream American English (Leonard, 2014; Redmond, et al., 2019; Rice, et al., 2004), as a strong clinical marker of language impairment. Although some children with DLD exhibit phonological deficits, they are not a defining feature. NWR difficulties have also been explored as clinical markers of language impairment, but some evidence suggests they are genetically separable from grammatical difficulties (Bishop, et al., 2006) and may be less sensitive and specific indicators of language impairment than sentence repetition tasks, which capture grammatical skills as well as working memory (Archibald & Joannise, 2009; Redmond, et al., 2011). The impaired language processes of children with DLD may also be more broadly related to memory deficits (Archibald & Gathercole, 2006; Bishop, et al., 2017; Edwards, Beckman, & Munson, 2004; Gathercole, et al., 1997; Gray, et al., 2019).

Numerous studies have established that dyslexia and DLD are separate disorders that often co-occur. From these studies, it is clear that children with co-occurring DLD and dyslexia (DLD+dyslexia) are poor at NWR, but the status of children with dyslexia-only and DLD-only is less clear. Catts et al. (2005) was the first study to establish classification criteria and examine NWR in separate versus co-occurring dyslexia and/or DLD. Tables 1 and 2 summarize details from Catts et al. (2005) and more recent studies that examined NWR in children with separate versus co-occurring dyslexia and DLD as they compare to peers with typical development (TD). Note that these studies were conducted prior to CATALISE and used SLI terminology and criteria. Table 1 presents the methodological details of the reported studies. Table 2 presents the results of various group comparisons on NWR. Numbers in the following summary correspond to the analysis rows in Table 2.

Table 1.

Descriptive information of studies investigating NWR performance in children with separate or co-occurring dyslexia and developmental language disorder

	Baird, Slonims, Simonoff, & Dworzynski (2011)	Bishop, McDonald, Bird, & Hayiou-Thomas (2009)	Catts, Adlof, Hogan, & Weismer (2005)	Fraser, Goswami, & Conti-Ramsden (2010)	McArthur & Castles (2013)	Ramus, Marshall, Rosen, & van der Lely (2013)	Rispens & Baker (2012) b	Rispens & Parigger (2010) b
Age/Grade of group classification	SLI and dyslexia between ages 6–13 years	SLI and dyslexia at 9 years	SLI at Kindergarten Dyslexia at 4th grade	Between 9–11 years	SLI and dyslexia between ages 7 −12 years	SLI and dyslexia between ages 8–12 years; TD between ages 5–12 years	SLI and dyslexia between ages 5.33–9.58 years	SLI and dyslexia between 7.75–8.83 years
Age/Grade of NWR measurement	Same as group classification	4 years; 6 years; 9 years	2nd grade; 8th grade	Same as group classification	Same as group classification	Same as group classification	Same as group classification	Same as group classification
TD (N)	49	176	165	13	8–9 a	65	41 (CA-TD) 16 (LA-TD)	15
Dyslexia-only (N)	N/A	73	21	14	41–43 a	21	14	N/A
SLI-only (N)	25	35	43	16	4–5 a	13	10	11
SLI+dyslexia (N)	19 (Moderate) & 21 (Severe)	54	18	21	17–19 a	30	23	18
NWR measurement task	Children’s Test of Nonword Repetition (Gathercole, Willis, Baddeley, & Emslie, 1994)	Children’s Test of Nonword Repetition (Gathercole, Willis, Baddeley, & Emslie, 1994)	Experimental (Dollaghan & Campbell, 1998)	Working Memory Test Battery for Children (WMTBC; Pickering and Gathercole, 2001)	NEPSY Repeating Nonwords subtest (Korkman, Kirk & Kemp, 1998)	Experimental (Ramus, Marshall, Rosen, and van der Lely, 2013)	Experimental (Rispens & Baker, 2012)	Experimental (De Bree, 2007)
NWR task scoring	Correct vs. Incorrect	Correct vs. Incorrect	Percent Consonants Correct (PCC)	Correct vs. Incorrect	Correct vs. Incorrect	Correct vs. Incorrect	Percent Phonemes Correct (PPC)	Percent Phonemes Correct (PPC)

Note. All studies referenced in this table were conducted before the CATALISE statements were published (Bishop, et al., 2016, 2017) and used the term “specific language impairment” (SLI). Across these studies, children were required to have average nonverbal IQ scores (≥ 80^th percentile). Studies differed in the number of groups included. “N/A” indicates when a specific group was not included in the study.

^aIn the McArthur & Castles (2013) study, four analyses were completed with differing group classifications: 1) receptive-expressive language + letter-sound reading, 2) receptive-expressive language + whole-word reading, 3) receptive language + letter-sound reading, and 4) receptive language + whole-word reading. The sample size varied across these different analyses.

^bChildren with SLI were recruited from schools and language impairment was verified by measurement of MLU.

Table 2.

Summary of group comparisons in studies investigating NWR performance in children with separate or co-occurring dyslexia and developmental language disorder

		Baird, Slonims, Simonoff, & Dworzynski (2011)	Bishop, McDonald, Bird, & Hayiou-Thomas (2009)	Catts, Adlof, Hogan, & Weismer (2005)	Fraser, Goswami, & Conti-Ramsden (2010)	McArthur & Castles (2013)	Ramus, Marshall, Rosen, & van der Lely (2013)	Rispens & Baker (2012)	Rispens & Parigger (2010)
1	TD vs. Dyslexia-only	N/A	Age 4 & 6: Not analyzed Age 9: d = 0.52, p < 0.05	p < 0.001	p < 0.05	A1- A4: N.S.	N.S.	N.S.	N/A
2	TD vs. SLI-only	Not analyzed	Age 4 & 6: Not analyzed Age 9: N.S.	p < 0.01	p < 0.05	Not analyzed	p ≤ 0.05	N.S.	N.S.
3	TD vs. SLI+dyslexia	Not analyzed	Age 4 & 6: Not analyzed Age 9: d = 1.16, p < 0.05	p < 0.001	p < 0.05	A1-A4: d = 1.13–2.66, p < 0.05	p ≤ 0.05	Chronological Age (CA) TD p < 0.01 Language Age (LA) TD p < 0.01	p = 0.013
4	Dyslexia-only vs. SLI+dyslexia	N/A	Age 4 & 6: Not analyzed Age 9: d = 0.65, p < 0.05	N.S.	N.S.	A1, A2, A4: d = 0.59–1.37, p < 0.05 A3: N.S.	p ≤ 0.05	p < 0.01	N/A
5	SLI-only vs. SLI+dyslexia	Moderate p = 0.007 Severe p = 0.004	Age 4 & 6: N.S. Age 9: d = 0.88, p < 0.05	p < 0.001	p < 0.05	Not analyzed	p ≤ 0.05	p < 0.05	p = 0.04
6	SLI-only vs. Dyslexia-only	N/A	Age 4 & 6: Not analyzed Age 9: N.S.	p < 0.001	N.S.	Not analyzed	N.S.	N.S.	N/A

Note. Numbered rows refer to the group comparisons summarized on p. 7 of the manuscript. Refer to Table 1 for the methodological details of the reported studies. The literature reviewed did not always include or analyze the same groups and information. The following are phrases and their definitions: a) “N/A” = Not applicable as the study did not include the group; b) “Not analyzed” = The group was included in the sample but not in reported statistical analyses; and c) “N.S.” = The group differences did not reach statistical significance. In the McArthur & Castles (2013) study, there were four analyses completed with differing group classifications: A1) receptive-expressive language + letter-sound reading, A2) receptive-expressive language + whole-word reading, A3) receptive language + letter-sound reading, and A4) receptive language + whole-word reading.

1. Three out of six studies found that children with dyslexia-only perform significantly worse than TD peers.

2. Three out of six studies found that children with DLD-only perform significantly worse than TD peers.

3. Seven out of seven studies found that children with DLD+dyslexia perform significantly worse than TD peers.

4. Four out of six studies found that children with DLD+dyslexia perform significantly worse than children with dyslexia-only.

5. Seven out of seven studies found that children with DLD+dyslexia perform significantly worse than children with DLD-only.

6. One out of five studies found that children with dyslexia-only performed more poorly on NWR than the DLD-only group.

Overall, there is clear evidence that NWR is most impaired in children with DLD+dyslexia relative to all other groups (see #3–5 above). These results also suggest that both single deficit groups demonstrated impaired NWR relative to TD (#1 and #2). However, the status of NWR in dyslexia-only versus DLD-only groups is less clear (#6). One possibility is that different underlying deficits in the dyslexia-only and DLD-only groups lead to similar levels of performance on NWR for these groups. For example, poorer NWR in children with dyslexia-only could be related to poorer phonological processing skills (Catts, et al., 2005), whereas in children with DLD-only, working memory or other lexical factors could play a role (Edwards, Beckman, & Munson, 2004; Gathercole, et al., 1997). Moreover, different underlying deficits could compound, leading to the greater impairment in children with DLD+dyslexia. For example, an underspecified phonological representation (due to dyslexia) may place a greater burden on reduced working memory resources (due to DLD), resulting in a greater deficit in performance relative to a single deficit. Alternatively, dyslexia-only and DLD-only groups may have similar processing difficulties during NWR that are additive when co-occurring.

An overarching goal of the current study was to examine NWR performance among these four groups of children to clarify group differences, especially for dyslexia-only and DLD-only groups. Most previous studies did not find significant differences in overall NWR performance between the single deficit groups. Thus, we probed the nature of NWR deficits using an experimental task designed to examine factors that may influence NWR performance. Presentation modality (i.e. audio-only or audiovisual) was manipulated to examine internal phonological representations. We also manipulated the phonemic variability of the nonwords (i.e., the amount of dissimilarity between articulatory features in the consonant phonemes within a word) to examine working memory processes.

Audiovisual Speech Processing in Dyslexia and DLD

Many studies of NWR present nonwords from auditory recordings or with the speaker’s mouth hidden (e.g., Bishop, Adams, & Norbury, 2004; Dollaghan & Biber, 1993; Gathercole, Willis, Baddeley & Emslie, 1994; Munson, Kurtz, & Windsor, 2005). However, in natural spoken communication, the speech signal is multi-modal, and the visual speech signal provides important phonological information, including place of articulation (Rosenblum, 2002). For example, the phonemes /f/ and /θ/ are acoustically similar, but visually different, such that disambiguating the two phonemes is relatively easier in face-to-face communication. Thus, probing the use of visual speech information may also provide information regarding internal phonological representation and processing.

Adults with typical development demonstrate strong benefits of visual speech information during speech-related tasks (e.g., Sumby and Pollock, 1954). However, individuals with language-based learning disabilities, such as dyslexia or DLD, show reduced benefit from visual speech information and show difficulties with audiovisual speech processing (deGelder & Vroomen, 1998; Francisco, Jesse, Groen, & McQueen, 2017; Leybaert, Macchi, Huyse, Champoux, Bayard, Colin, & Berthommier, 2014; Megnin-Viggars & Goswami, 2013; Meronen, Tiippana, Westerholm, & Ahonen, 2013; Norrix, Plante, Vance, & Boliek, 2007; Ramirez & Mann, 2005). Furthermore, adults with dyslexia have reduced neural enhancement for speech recognition with visual cues compared to adults with strong word reading skills (Ye, Rüsseler, Gerth, & Muente, 2017).

Taken together, these studies suggest that individuals with dyslexia or DLD may have poorer audiovisual speech processing relative to individuals with typical language and reading skills, potentially due to degraded phonological representations. However, none of these studies considered the co-occurrence of dyslexia and DLD, which complicates group comparisons. Investigating the separate and overlapping deficits among these groups is necessary for delineating potential differences in the nature of their phonological representations.

Working Memory in Dyslexia and DLD

In addition to differences in phonological representations, it is possible that individuals with dyslexia and/or DLD differ in how those nonwords are encoded and processed within working memory. Working memory is a temporary storage for maintaining perceptual information (or information retrieved from long-term memory) in conscious awareness (e.g., Baddeley, 1992). However, the similarity of acoustic information held in memory can significantly affect recall performance. For example, memory performance is better for lists of consonants (and vowels) that are highly discriminable across manner classes, such as /g, ʃ, m/, compared to highly similar consonants within the same manner class, such as /b, d, g/ (Darwin & Baddeley, 1974). This is because acoustically similar phonemes are more likely to be confused during maintenance in working memory, particularly as the stored representation of the nonword degrades over time. This degradation can be further compounded if the phonological representation is underspecified to begin with, as thought to be a key feature of dyslexia. Maintaining distinct representations of the phonemes to be recalled in memory is likely to be more difficult with increasing similarity of the individual phonemes within the nonword to be recalled. Here, we manipulated phonemic variability, or the dissimilarity between individual consonants in the nonword; thereby, providing new insights into working memory limitations and differences among dyslexia and DLD groups.

Purpose of Study

The purpose of the current study was to examine differences in NWR performance between children with dyslexia-only, DLD-only, DLD+dyslexia, and TD. We examined these differences with a commonly used, norm-referenced assessment of NWR and an experimental NWR task. For the norm-referenced task, we hypothesized that the TD group would achieve higher scores than the DLD+dyslexia group. However, given mixed results of past studies, we made no hypothesis about whether the single disorder groups (dyslexia-only or DLD-only) would differ significantly from the TD group or from one another.

The experimental task was designed to examine potential processing differences among the groups through the manipulation of phonemic variability (high or low) and presentation modality (audio-only or audiovisual). Phonemic variability refers to the level of dissimilarity between the articulatory features of the consonant phonemes within the nonword. Nonwords with more similar consonant phonemes had a low phonemic variability, whereas nonwords with more distinct phonemes had a high phonemic variability. The manipulation of phonemic variability was designed to assess memory processes during maintenance in working memory. Here poorer memory processes should lead to greater degradation in memory of the nonword, increasing the likelihood of confusions during maintenance. For TD children, we predicted that NWR accuracy would be greater for words presented in the audiovisual condition, where more phonological information was available, than in the audio-only condition. Further, we predicted this effect would be more pronounced for low phonemic variability nonwords, for which constituent consonants were differentiated only by place of articulation. In this condition the presentation of visual articulatory place features should enhance the encoding of phonological representations (Baird, et al., 2011). Finally, if dyslexia-only and DLD-only groups differ in their source of NWR difficulty (e.g., representation versus memory, respectively), we hypothesized that these groups would also show differential effects of phonemic variability and presentation modality.

Method

All procedures in this study were approved by the University of South Carolina Institutional Review Board. Parents of child participants provided signed informed consent, and the children gave verbal assent prior to their participation in the study.

Participants

A total of 160 second grade children, ages 7–9 years, participated as a part of a larger project conducted in South Carolina, USA (Adlof, Scoggins, Brazendale, Babb, & Petscher, 2017; Adlof, Baron, Bell, & Scoggins, in review). Children were recruited through a classroom screening process (Adlof et al., 2017) and classified into four groups based on their performance on a battery of individually administered language and reading assessments. Inclusion criteria were as follows: children with TD (N = 59) scored between the 25th and 85th percentiles on the Woodcock Reading Mastery Test-3 Basic Skills composite, which includes subtests of real word and pseudoword reading (WRMT-3; Woodcock, 2011), and the Clinical Evaluation of Language Fundamentals – 4 Core Language Score (CELF-4 CLS; Semel, Wiig, & Secord, 2006). The CLS is derived from four subtests (i.e., Concepts and Following Directions, Word Structure, Recalling Sentences, and Formulated Sentences), which assessed receptive and expressive aspects of oral language at the word and sentence level. Children with dyslexia-only (N = 20) scored below the 16^th percentile on the WRMT-3 and above the 25^th percentile on the CELF-4. Children with DLD-only (N = 40) scored above the 25^th percentile on the WRMT-3 and below the 16^th percentile on the CELF-4. Children with DLD+dyslexia (N = 41) scored below the 16^th percentile on both the WRMT-3 and CELF-4. Table 3 lists the results of these classification assessments by group.

Table 3.

Group means, standard deviations, and significance of pairwise comparisons for group classification assessments

	Dyslexia-only (N = 20)	DLD-only (N = 40)	DLD+dyslexia (N = 41)	TD (N = 59)
CELF-4 Core Composite Standard Score	97.95a (5.84)	78.70 b (5.54)	73.44 c (7.54)	100.68a (6.89)
WRMT-III Basic Skills Standard Score	80.65 a (3.72)	101.43b (7.58)	74.49 c (6.57)	100.86b (6.33)
WRMT-III Word Identification Standard Score	84.75a (5.55)	101.58b (7.82)	78.41c (7.56)	103.58b (6.87)
WRMT-III Word Attack Standard Score	78.90a (7.68)	101.48b (8.41)	73.10c (7.55)	98.85b (7.95)

Note. Planned pairwise comparisons were conducted using Tukey HSD test. Groups that share subscripts are not significantly different (p < 0.05). Bolded group means and standard deviations indicate the composites demonstrating an impairment used to classify each group.

Because participants were tested in an area of the country where nonmainstream dialects of English (NMAE) are relatively common, it was important to ensure that children who spoke an NMAE dialect were not incorrectly classified as language impaired by the CELF-4, which assesses comprehension and production of Mainstream American English (MAE). The Diagnostic Evaluation of Language Variation-Screening Test (DELV-ST; Seymour, Roeper, & de Villiers, 2003) was used to determine which students spoke NMAE versus MAE and to confirm that students classified as DLD or DLD+dyslexia on the CELF-4 also showed a high risk for language impairment on the DELV-ST, which is a dialect-neutral assessment (cf. Hendricks & Adlof, 2017).

Participants also completed additional language and cognitive measures to as a part of the larger project. These descriptive measures included the Test of Nonverbal Intelligence-4 (TONI-4; Brown, Sherbenou, & Johnsen, 2010), the Peabody Picture Vocabulary Test-4 (PPVT-4; Dunn & Dunn, 2007), the Test of Word Reading Efficiency-2 (TOWRE-2; Torgesen, Wagner, & Rashotte, 2012), the Expressive Vocabulary Test-2 (EVT-2; Williams, 2007), and the Memory for Digits subtest of the Comprehensive Test of Phonological Processing-2 (CTOPP-2; Wagner, Torgesen, Rashotte, & Pearson, 2003). Table 4 provides a summary of group performance.

Table 4.

Group means, standard deviations, and significance of pairwise comparisons for group descriptive assessments and NWR tasks

	Dyslexia-only (N = 20)	DLD-only (N = 40)	DLD+dyslexia (N = 41)	TD (N = 59)
Descriptive Assessments
TOWRE-2 Scaled Score	79.75a (7.33)	101.28b* (10.16)	72.90a (9.33)	101.73b (10.67)
TOWRE-2 Sight Word Efficiency Scaled Score	85.70a (12.37)	104.46b* (9.39)	79.41a (11.10)	105.59b (10.63)
TOWRE-2 Phonemic Decoding Efficiency Scaled Score	75.85a (6.30)	97.87b* (11.58)	67.54c (12.54)	96.92b (10.35)
PPVT-4 Standard Score	106.35a (8.24)	92.30b (7.66)	89.88b (10.87)	106.52a* (10.34)
EVT-2 Standard Score	101.40a (6.15)	92.43b (7.28)	85.66c (7.64)	106.41a (8.21)
TONI-4 Index Score	101.45a,c (10.67)	100.49a* (9.90)	94.63b (8.12)	106.66c (8.82)
CTOPP-2 Memory for Digits Scaled Score	9.60a,c (2.96)	9.32a,d (2.14)	7.93b (2.08)	10.36c,d (2.38)
Nonword Repetition (NWR) Tasks
CTOPP-2 Nonword Repetition Scaled Score	5.95a,c (1.99)	5.53a (1.76)	4.07b (2.07)	7.08c (2.25)
Overall Experimental Nonword Repetition Task Score	19.30a,b (5.94)	20.18a (6.85)	15.20b (6.92)	25.15c (5.63)

^(*)Note. Planned pairwise comparisons were conducted using Tukey HSD test. Groups that share subscripts are not significantly different (p < 0.05). One participant did not receive the norm-referenced assessment. Sample number change indicated by.

Norm-referenced Nonword Repetition Task

Participants completed the Nonword Repetition subtest of the CTOPP-2, which consists of 30 recorded nonwords ranging from 1 to 9 syllables (3 – 15 phonemes) in length. The nonwords were scored dichotomously as correct if they were exact repetitions or incorrect if there were any differences between the child’s production and the recorded stimulus. Items were administered until participants reached a ceiling of three consecutive incorrect items. Scaled scores were used for the analyses in this study.

Experimental Nonword Repetition Task

Stimuli.

The nonword stimuli were video recorded in a sound attenuated booth by a female native English speaker. These recordings were used for both audio-only (AO) and audiovisual (AV) presentations. The nonword stimulus set included 36, three-syllable nonwords with CVCVCV structure. The nonwords were constructed with combinations of three vowels, /i, ɑ, u/, and nine voiced consonants, /b, d, g, m, n, ŋ, v, z, ʒ/ from three manner classes (stops, nasals, and fricatives). In half of the nonwords, the three consonants within each word were all from the same manner class but had different places of articulation (e.g., /bigudɑ/). Because only one articulatory feature (i.e., place) differed between the consonants, which had the same manner and voicing, these nonwords were described as having low phonemic variability. The other half of the nonwords varied both the manner and place of articulation across the three consonants (e.g., /zigumɑ/). These nonwords were described as having high phonemic variability. No syllables or phonemes were repeated within any nonwords. The Appendix includes the full list of nonword stimuli. The internal consistency of the 36 nonword items in this task was high with this participant sample (α = 0.883).

Procedure.

The task was presented to participants using E-Prime 2.0 software (Schneider & Zuccoloto, 2007). Half of the nonword stimuli were randomly assigned to the AO modality, and the other half were assigned to the AV modality. Nonwords were presented in four blocks of nine words, with the order of AO and AV block presentations counterbalanced across participants. The order of nonwords within each block was randomized for each participant. Participants were seated directly in front of the screen. They were instructed to listen to each nonword and repeat back exactly what they heard. The examiner controlled the presentation of nonwords; each new nonword was presented immediately after the previous nonword had been repeated. A short break was allowed if requested. No feedback about accuracy was provided. Participants’ spoken responses were video- and audio-recorded for later transcription and scoring.

Scoring.

Scorers were trained and assessed on broad phonetic transcription and scoring, and each scorer was required to pass a reliability test before scoring independently. Scorers phonetically transcribed each nonword production from the video recording. In a few cases, the audio recording was used if there was an error during video recording. Correctly produced phonemes were those that were repeated accurately in the correct order. The percent of phonemes correctly produced (PPC) was calculated by adding up the total number of correctly produced phonemes and dividing by the sum of the total number of phonemes in the target nonword and any incorrectly inserted sounds.

Reliability.

A random sample of 20% of the experimental NWR transcriptions were doubly transcribed and scored from the recordings in order to measure transcription and scoring reliability, which were calculated separately. For transcription reliability, percent phoneme agreement was calculated by adding up the total number of phonemes agreed upon and dividing it by the total number of phonemes transcribed. The percent phoneme agreement between scorers was 95.9%. For scoring reliability, overall agreement was calculated by adding up the total number of items agreed upon and dividing by the total number of items. The item-by-item agreement between scorers was 90.6%.

Results

Descriptive statistics are provided in Table 3 for the norm-referenced assessments used to classify the groups, and in Table 4 for the descriptive assessments and the NWR tasks. Analysis of variance (ANOVA) was used to examine group differences. Significant between-group differences were found for all classification assessments (all Fs > 127.84, all ps < 0.001; see Table 3) and all descriptive assessments (all Fs > 9.01, all ps < 0.001; see Table 4). Overall, the descriptive assessments support the validity of the subgroup classifications. In addition to scoring low on the classification measures of word reading accuracy, the dyslexia-only and DLD+dyslexia groups showed significantly poorer performance on descriptive measures of word- and nonword-reading fluency than the DLD-only and TD groups. Likewise, in addition to scoring low on the omnibus language assessment used to classify the groups, the DLD-only and DLD+dyslexia groups showed significantly poorer performance on measures of receptive and expressive vocabulary than the dyslexia-only and TD groups.

In terms of phonological memory as measured by the CTOPP-2 Memory for Digits subtest, the DLD+dyslexia group had poorer phonological memory than their peers with a single disorder (dyslexia-only and DLD-only) and peers with TD. The single disorder groups (dyslexia-only and DLD-only) were found to have similar phonological memory abilities, which fell in between those of the DLD+dyslexia and TD groups, although only the DLD-only group significantly differed from the TD group. Finally, although we did not exclude children from participation on the basis of nonverbal IQ scores, the dyslexia-only and DLD-only groups scored near the population mean, and most participants in all groups scored within one standard deviation of the mean (Dyslexia-only = 90.0%, DLD-only = 89.7%, DLD+Dyslexia = 90.2%, TD = 100.0%). Still, there were significant differences in nonverbal IQ performance between groups, a finding consistent with other studies of SLI (Gallinat & Spaulding, 2014). Thus, the sample of children with DLD in this study is largely comparable to previous studies of SLI.

To address our research questions, we first examined group differences on the norm-referenced assessment of NWR, the CTOPP-2 subtest. Figure 1 illustrates the distribution of scores for each group. One-way ANOVA results showed a significant main effect of group [F (3, 156) = 19.03, p < 0.001; η_p² = 0.27]. We conducted planned follow-up analyses with Tukey’s HSD and measured the effect size of group differences with Hedges’ g. The TD group outperformed the DLD-only (p = 0.001; g = 0.76) and DLD+dyslexia groups (p < 0.001; g = 1.39), but not the dyslexia-only group (p = 0.141; g = 0.45). The dyslexia-only and DLD-only groups did not significantly differ from one another (p = 0.865; g = 0.31), and both scored significantly higher than the DLD+dyslexia group (DLD-only vs. DLD+dyslexia, p = 0.005; g = 0.76; dyslexia-only vs. DLD+dyslexia, p = 0.004; g = 0.98).

Figure 1.

Boxplots illustrating group differences on the CTOPP-2 NWR subtest

Next, we examined the effects of group (dyslexia-only, DLD-only, DLD+dyslexia, and TD), phonemic variability (high vs. low), and presentation modality (AO vs. AV) on NWR performance in the experimental task. Initial data screening indicated that NWR accuracy, measured by PPC, was relatively high across conditions, and the distributions of scores were negatively skewed. Therefore, in order to standardize the error variance, we transformed the data using a rationalized arcsine transformation (Studebaker, 1985). We conducted a within- and between-groups mixed ANOVA for both the transformed and untransformed data. The pattern of statistically significant results was the same between the two analyses; therefore, we report the untransformed scores for ease of interpretation. Group comparisons were also analyzed using dichotomous scores with no differences from the PPC results; therefore, we report the PPC results.

Figure 2 illustrates the mean PPC by phonemic variability level and presentation modality for each group. Results showed significant main effects of group [F (3, 156) = 20.39, p < 0.001; η_p² = 0.28] and phonemic variability level [F (1, 156) = 63.83, p < 0.001, η_p² = 0.29]. The main effect of modality was not significant [F (1, 156) = 2.16, p = 0.144; η_p² = 0.01]. There was a significant two-way interaction between phonemic variability and modality [F (1, 156) = 15.15, p < 0.001, η_p² = 0.09]. The three-way interaction was not significant [F (3, 156) = 1.63, p = 0.186, η_p² = 0.03], and group did not significantly interact with phonemic variability level [F (3, 156) = 0.27, p = 0.848; η_p² = 0.01] or modality [F (3, 156) = 2.16, p = 0.144; η_p² = 0.01]. Follow-up analyses revealed that the TD group outperformed all other groups (all p < 0.005), with moderate effect sizes compared to dyslexia-only (g = 0.49) and DLD-only (g = 0.49) and a large effect size compared to DLD+dyslexia (g = 1.08). The DLD-only group had significantly higher scores than the DLD+dyslexia group, with a moderate effect size (p < 0.001; g = 0.61). The effect size for the difference between dyslexia-only and DLD+dyslexia was also moderate but did not reach significance (p = 0.063; g = 0.43), and the dyslexia-only and DLD-only groups did not differ from each other (p = 0.919; g = 0.11).

Figure 2.

Percent phonemes correct (PPC) for each phonemic variability level (low on the left and high on the right) by presentation modality for each group. AO = auditory-only; AV = audiovisual. Error bars = 95% confidence interval.

To decompose the significant two-way interaction, we examined the effect of modality within each level of phonemic variability separately including group as a factor. The effect of modality was significant for low phonemic variability nonwords [F (1, 156) = 14.38, p < 0.001; η_p² = 0.08], where performance was better in the AV modality than the AO modality. The effect of modality was not significant for the high phonemic variability nonwords [F (1, 156) = 2.21, p = 0.14; η_p² = 0.01]. However, the effect of variability was significant within both modalities [AO: F (1, 156) = 85.53, p < 0.001; η_p² = 0.35; AV: F (1, 156) = 9.10, p = 0.003; η_p² = 0.06]. These results suggest that AV presentation facilitated NWR when the phonemes within the nonword were perceptually similar (i.e., low phonemic variability), but not when they were highly discriminable (i.e., high phonemic variability).

Although not significant, visual inspection of Figure 2 suggests two trends that we examined post hoc. First, AV presentation was associated with better performance in low phonemic variability nonwords, but performance appeared to be slightly lower in the AV condition than the AO condition for high phonemic variability nonwords. The low phonemic variability nonwords involved consonants at three different places, whereas high phonemic variability nonwords involved two different places for seven out of 18 nonwords (the remaining 11 having three places; see the Appendix). Post-hoc analyses indicated that NWR performance was better for high phonemic variability nonwords with both two- and three-consonant places of articulation compared to low phonemic variability nonwords. Taken together, these support the conclusion that high phonemic variability within the nonwords facilitated their repetition such that visual articulatory information provided no further benefit.

Second, despite a lack of significant group interactions, visual inspection of Figure 2 suggests a potentially different trend for children with dyslexia-only compared to the other groups. To explore this possibility, two sets of difference scores were examined. First, differences between the two presentation modalities (i.e., AV – AO = modality difference score) were examined for the low phonemic variability condition in which visual cues had the greatest effect. Second, differences between the two phonemic variability levels (i.e., high – low = variability difference score) were assessed for the AO condition in which effects of phonological similarity were most evident. A one-way ANOVA was run for each difference score. No significant effects of group were found, further supporting the similar performance for dyslexia-only and DLD-only on these NWR tasks.

Discussion

This study set out to examine differences in NWR between children with dyslexia-only, DLD-only, DLD+dyslexia, and TD. Comparisons of NWR involving dyslexia and DLD have been difficult due to the high co-occurrence of these disorders in children, which has not been controlled for in many past studies. Furthermore, traditional NWR tasks might not be designed to identify differences in the source of phonological processing difficulties among these groups. To address these challenges, we enlisted a relatively large sample of second-grade students (age 7–9 years) who met criteria for separate or co-occurring dyslexia and DLD, or TD. We compared group performance on a norm-referenced nonword repetition (NWR) task and an experimental NWR task manipulating phonemic variability and presentation modality to probe potential phonological processing differences among the groups. The manipulation of modality altered the phonological information available for children to form robust phonological representations of the presented nonword. In contrast, the manipulation of phonemic variability determined the dissimilarity of the phonemic elements of the nonword needing to be stored and recalled from phonological working memory, thus testing the integrity of the system to store robust and separate phonological representations of the nonword consonants in memory.

In the experimental task, phonemic variability significantly influenced NWR performance in both AO and AV presentation modalities. Thus, phonemic variability within nonwords is a general stimulus characteristic that influences NWR performance and can be considered when evaluating NWR tasks and performance. Additionally, the benefit of AV presentation over AO presentation was only significant for nonwords with low phonemic variability. The addition of highly discriminable visual cues may have facilitated maintenance of robust phonological representations for repeating nonwords that have acoustically similar phonemes, differentiated by place of articulation. In contrast, the limited effect of visual cues for nonwords with high phonemic variability consonants may have occurred because either children were already performing at their ceiling during AO, or the visual phonemic information did not provide additional benefit over the already highly discriminable consonants that differed in both manner and place. More generally, when NWR tasks feature words with easily confusable phonemes (i.e., words with low phonemic variability), AV presentation can facilitate more accurate repetitions.

In the experimental task, there was a significant main effect of group, but no interaction with phonemic variability or presentation modality. Instead, the pattern of group performance was broadly consistent with results from the CTOPP-2 NWR task. Across both tasks, the dyslexia-only and DLD-only groups performed intermediately to the TD group and the DLD+dyslexia group, and these single-deficit groups did not significantly differ from each other. The lack of an interaction between group and phonemic variability or presentation modality in the experimental task suggests that these factors influence children with dyslexia and DLD in similar ways to each other and to children with TD.

Overall, the results from our experimental NWR task are in line with many previous studies, which have shown that adults and children with TD demonstrate strong benefits of visual speech information during speech-related tasks (deGelder & Vroomen, 1998; Leybaert, et al., 2014; Megnin-Viggars & Goswami, 2013; Meronen, et al., 2013; Norrix, et al., 2007; Ramirex & Mann, 2005; Sumby and Pollock, 1954). However, these studies demonstrated reduced audiovisual benefit for individuals with a current, or history of, language-based learning disability (i.e., dyslexia, DLD) compared to individuals with TD, whereas our study showed a similar benefit for all groups. Differences in stimuli (e.g. multisyllabic versus monosyllabic) may account for some of these differences. Also, notable differences in the size of our sample and the unique control for the co-occurrence of dyslexia and DLD may explain these discrepancies.

Group comparisons from both the norm-referenced and experimental NWR tasks were broadly consistent with the majority of past studies that did not find a significant difference in NWR performance between dyslexia-only and DLD-only groups, but did find that children with DLD+dyslexia showed the poorest levels of performance (see Table 2). When the disorders co-occur, the children with TD, as well as those with only a single disorder, outperform children with DLD+dyslexia. This group effect was fairly consistent across the norm-referenced task and all conditions in the experimental task, although the difference between dyslexia-only and DLD+dyslexia did not reach significance in the experimental task.

Past studies had also reported mixed results with regard to the differences between the single deficit groups and TD (see Table 2). Many of the previous studies involved participants from a wide age range, who presumably also differed in reading ability and reading experience, which may have obscured NWR differences related to group. NWR has been shown to be a strong predictor of reading ability early on, but reading ability predicts NWR performance at older ages (Nation & Hulme, 2011). A strength of our study is its relatively large sample size with a narrow age range. We found significant, moderate differences between the single deficit groups and TD in both the norm-referenced and the experimental NWR tasks. Importantly, these results suggest that both dyslexia and DLD alter the phonological processing necessary for typical NWR.

We did not find evidence of qualitative differences in phonological processing for NWR between the single deficit groups in this study, and effect sizes for quantitative differences were quite small. Either these two groups have similar sources of NWR difficulty, or additional studies will need to probe phonological representation and processing in these two groups further. For example, there may be a difference in their phonological processing that is not captured by measuring accuracy of NWR. Error coding methods may be able to identify qualitative differences between the types of NWR errors made between the dyslexia-only and DLD-only groups. Similarly, other tasks besides NWR may be more ideal for specifically examining different types of phonological processing separately. Another possibility is that dyslexia-only and/or DLD-only exhibit initially similar phonological deficits which are differentially impacted by other language skills in addition to reading instruction and experience. Addressing such questions requires longitudinal data collected before and after the onset of formal reading instruction. Although no studies in our review had NWR data collected from all four groups before and after formal schooling, findings from Bishop et al. (2009) and Catts et al. (2005) suggest that dyslexia-only, DLD-only, and/or DLD+dyslexia showed similar levels of phonological deficits in preschool and kindergarten but DLD-only showed improvement over time. Our study found that children with DLD-only showed similar levels of NWR difficulty to dyslexia-only, but this may be related to the fact that we identified DLD and dyslexia at the same time point that we measured NWR, whereas Catts et al. (2005) identified DLD in kindergarten and dyslexia in fourth grade.

While there are clear quantitative differences between dyslexia-only and DLD-only groups with regard to word reading and broader language skills (e.g., morphology, semantics, syntax), these skills are intercorrelated and the boundaries between groups are arbitrarily drawn. Furthermore, it is now more commonly acknowledged that both of these disorders are multifactorial (Astle & Fletcher-Watson, 2020; Catts, McIlraith, Bridges, & Nielsen, 2017), and phonological processing deficits alone are insufficient to explain either dyslexia or DLD. Importantly, when the two disorder co-occur, NWR performance was poorer than for either disorder alone. The effect sizes for our DLD+dyslexia group compared to the TD group in the norm-referenced (g = 1.39), and experimental NWR task (g = 1.08) were similar in size to the average effect sizes reported in the meta-analyses for children with dyslexia (d = 0.92 ; Melby-Lervag & Lervag, 2012) and SLI (d = 1.27; Graf-Estes, et al., 2007). In contrast, the effect sizes for single deficit groups were approximately half of those for the co-occurring group. This compounded effect provides further argument for the need to detail the co-occurrence of these disorders, which is missing from many studies.

Compared to previous studies, here we report findings from a relatively large sample size of children that includes participants from all subgroups, allowing us to explore and potentially delineate differences in the phonological processing unique to either dyslexia, DLD, or the combination. Despite these strengths, we acknowledge two limitations. First, the dyslexia-only group was smaller than other groups we tested. However, even if statistical power was somewhat limited for this group, the pattern of effect sizes does not suggest a different interpretation of the results than the statistical significance tests indicate. Second, this study did not vary the length or difficulty of nonwords, which may help to track performance levels at which the groups are maximally distinguishable. Based on previous studies we selected three-syllable nonwords that have been determined to be the most difficult for children with dyslexia and DLD (Graf-Estes, et al., 2007; Melby-Lervag & Lervag, 2012), predicting that this level of difficulty would facilitate the determination of group differences. However, in our study all children performed relatively well, perhaps contributing to the lack of significant NWR performance differences between the single-disorder groups.

Conclusion

The current study highlights differences in NWR performance for children with separate versus co-occurring dyslexia and DLD. First, we found that children with DLD and dyslexia both showed significant deficits in NWR relative to TD children. Second, we found that the DLD+dyslexia group performed worse than the single-disorder groups, suggesting that the deficits related to each disorder may compound, leading to increased errors in NWR. However, we did not find significant differences in NWR accuracy between dyslexia-only and DLD-only groups. In the experimental task, audiovisual presentation facilitated accurate repetition of low phonemic variability nonwords, but this effect was consistent across groups. Thus, the current analysis does not provide evidence for a different underlying cause of NWR difficulties in dyslexia-only versus DLD-only groups. Future studies examining the types of errors made by dyslexia-only or DLD-only groups may provide further insight into the existence of any qualitative differences in phonological processing.