Extending Complexity to Word-Final Position via Telepractice: Intervention Effects for English-Speaking Children With Speech Sound Disorder

Abstract

Purpose:

Phonologically complex targets (e.g., [pl–]) are understood to facilitate widespread gains following speech sound intervention, and yet, available research largely features word-initial clusters. The present study investigates intervention effects following treatment of complex clusters presented in word-final position. Importantly, this allows for an added layer of complexity via suffixes that mark tense and agreement.

Method:

Eight English-speaking children with speech sound disorder (SSD; 3;3–6;9 [years;months]) participated in 18 one-on-one intervention sessions. Intervention was completed via telepractice, as were all pre- and post-intervention assessments. Intervention targets were word-final two-element consonant clusters that were unknown to the child prior to intervention. Targets were presented in verbs that were either monomorphemic (i.e., [−ks]; they mix) or bimorphemic (i.e., [−ks]; she pick/3s, marked for third-person singular).

Results:

All participants demonstrated change across multiple phonological measures. More stringently, six of eight participants demonstrated generalization to untreated sounds and untreated words immediately following intervention, including four of four children with monomorphemic targets. Importantly, positive changes for children with both target types were observed following a relatively short course of intervention (18 sessions over 6 weeks), and mastery of the target cluster was not required for phonological growth to occur.

Conclusions:

Results align with available work featuring word-initial complex targets and indicate that word-final consonant clusters are feasible, effective targets for English-speaking children with SSD. Findings similarly affirm the use of telepractice to deliver research-based interventions. Speech-language pathologists may thus integrate these findings with their clinical judgment and client perspectives to implement such targets in clinical practice.

Functional speech sound disorder (SSD) is often identified in early childhood and is characterized by difficulties with perceiving and/or producing speech sounds. The impact of persistent SSD is long-lasting, with implications for social–emotional development, academic achievement, and vocational outcomes later in life (Lewis et al., 2015). The high prevalence of this disorder is reflected in the caseloads of practicing speech-language pathologists (SLPs). Per the 2022 Schools Survey Report by the American Speech-Language-Hearing Association (ASHA, 2022), 91.8% of SLPs regularly serve students with SSD, creating a demand for highly effective and efficient clinical intervention practices that support this population.

The complexity approach is an evidence-based target selection process for speech sound intervention that promotes widespread phonological change (Maggu et al., 2021). Under this paradigm, consonant clusters targets (e.g., two-element [pl–] in play) are more phonologically complex than singletons (e.g., [p–] in pay) and support generalization to untreated speech sounds (Barlow, 2005; Gierut, 2007). Importantly, previous studies have primarily evaluated targets in word-initial position. Early findings suggest that word-final consonant clusters (e.g., [−ks] in mix) are also effective treatment targets for children with SSD (Combiths, Barlow, Richard, & Pruitt-Lord, 2019; Potapova et al., 2023). Critically, word-final consonant clusters may also feature morphosyntactic elements in languages with suffixes, such as tense and agreement marking in English (e.g., third-person singular [/3s]; she pick/3s). As a result, this word position allows for manipulation of complexity across two domains: phonology and morphology. Intervention targets featuring such cross-domain complexity may increase positive treatment outcomes for children with SSD and reduce time spent in therapy (Potapova et al., 2023; Tyler et al., 2002). Furthermore, word-final phonological processes (e.g., omissions, substitutions) are often observed in children with SSD (e.g., Weismer et al., 1981), adding to the motivation for the present study. The current work aims to expand the field's understanding of the effects of word-final consonant cluster targets, with and without morphological complexity, to better equip clinicians in supporting children with SSD.

Evidence for intervention approaches is strengthened as multiple service delivery models are considered. Use of telepractice increased significantly over the course of the COVID-19 pandemic and continues to have a positive impact on the reliability, efficacy, cost-effectiveness, and accessibility of speech and language services (e.g., Kollia & Tsiamtsiouris, 2021; Molini-Avejonas et al., 2015). Available research affirms that telepractice and in-person speech sound interventions are comparable, though research specifically featuring complex targets is limited. Use of telepractice in the current study provides new data that extend the telepractice evidence to include consonant cluster targets in speech sound intervention (see John et al., 2023).

The Complexity Approach in SSD Intervention

Complexity-based accounts contend that treatment of complex structures that are linguistically and hierarchically related leads to generalization to untreated, less complex structures (Thompson, 2007). Generalization refers to the effect of treated intervention targets on structures not directly addressed in intervention. This outcome has been notably observed with phonologically complex treatment targets for children with SSD (e.g., Gierut et al., 1996; Miccio & Ingrisano, 2000). Specifically, complex speech sound targets, including consonant clusters, engage a child's linguistic system and prompt generalization to untreated, less complex speech sounds (see Gierut, 2007; Maggu et al., 2021). Accordingly, this approach is well suited—and well studied—for children with functional SSD who have multiple sounds missing from their inventories (e.g., Kamhi, 2006; Storkel, 2018). Speech sound targets within the complexity approach include two main features, as described by Storkel (2018) and summarized here.

First, the linguistic characteristics of a complex target account for implicational universals, or phonological patterns observed across languages that indicate the presence of complex sounds must co-occur with less complex sounds but not vice versa (Storkel, 2018). For example, research has demonstrated that the presence of consonant clusters implies the presence of consonant singletons within a language system; however, the presence of singletons does not imply clusters (Watts & Rose, 2020). Accordingly, clusters are relatively more complex than singletons. This universal is similarly observed for children with SSD (Gierut & Champion, 2001). The second main feature of complex speech sound targets is that they should be least known by the child, as indicated by low accuracy in spontaneous and imitated production (Storkel, 2018). Evidence has shown that treatment on least-known speech sounds supports generalization to untreated speech sounds for children with SSD (Gierut et al., 1987; Powell, 1991). Importantly, word-final phonological errors in children with SSD are prevalent (Weismer et al., 1981), indicating limited knowledge of speech sounds in that specific context. As such, word-final targets may be valuable in manipulating complexity.

Altogether, target complexity must be considered carefully. While least-known sounds typically facilitate greater generalization, there may be limits to how complex least-known targets can be (Gierut & Champion, 2001). Thus, optimal intervention targets are determined both by linguistic sources of complexity and individual differences, including phonological knowledge—and word-final targets allow for each of these factors to be considered in target selection.

Word-Final Complexity

Studies evaluating the complexity approach thus far have predominantly featured complex sounds in the word-initial position (e.g., Elbert et al., 1984; Gierut, 1999). Word-final treatment targets offer opportunities to manipulate complexity and individualize intervention. For example, word-final targets can include monomorphemic consonant clusters (e.g., [−ks] in mix) or bimorphemic clusters that mark tense and agreement (e.g., [−ks] in pick/3s features third-person singular). While both types of clusters are phonologically complex, monomorphemic clusters are relatively simple with respect to morphology, and bimorphemic clusters, carrying additional grammatical information, are relatively complex in morphology. Understanding the impact of leveraging one or more layers of linguistic complexity may offer new directions in maximizing intervention outcomes for children with SSD. Indeed, early research in this area indicates that targets featuring complexity across phonology and morphology have the potential to promote growth in a child's system.

Previous studies using cross-domain complexity targets have primarily featured intervention for children with co-occurring SSD and developmental language disorder (SSD–DLD). In a study by Tyler and Sandoval (1994), two children (ages 3;6 and 4;3 [years;months]) received treatment on word-final cluster targets (e.g., [−ts], cat/s; [−ps], jump/3s) that were chosen for their morphological and phonological properties (i.e., consonant clusters in plural and third-person singular forms, respectively). Both participants exhibited growth on phonological measures as well as generalization to untreated morphological forms, suggesting that treatment targets with multiple layers of complexity should be evaluated for potential positive impacts across both linguistic domains for children with SSD–DLD. Similarly, a case study by Combiths, Barlow, Richard, and Pruitt-Lord (2019) described the effects of treatment on a word-final consonant cluster target with third-person singular marking (e.g., [−lps] in help/3s) for Max, a child (5;2) with SSD–DLD. Phonological results following intervention indicated broad generalization to consonant singletons and clusters in word-initial and word-final positions. These initial findings suggest that word-final complex targets are appropriate for intervention in SSD in improving phonological outcomes, though more research is needed to establish their efficacy and reliability.

A case study by Potapova et al. (2023) extended the above work to examine consonant clusters in the word-final position with and without morphological complexity for children with SSD (and typically developing language). The first participant, Anna (3;7), received treatment on a monomorphemic (i.e., morphologically simple) word-final cluster ([−ks] in mix) and, post-intervention, produced consonant clusters previously missing from her repertoire and, per caregiver report, demonstrated increased intelligibility. The second participant, David (4;1), received treatment on a bimorphemic (i.e., morphologically complex) word-final cluster ([−ks] in pick/3s). Post-intervention, David demonstrated more robust gains than Anna, including generalization to untreated consonants and consonant clusters. This pattern provided early evidence that targets with multiple levels of complexity (i.e., phonology and morphology) may be particularly effective intervention targets. However, the highly individualized nature of SSD may have also contributed to variability across participants' outcomes. For example, David had relatively greater phonological knowledge (e.g., higher percent consonants correct; Shriberg et al., 1997) than Anna prior to intervention, suggesting that child-specific factors contribute to intervention outcomes (see also Gierut et al., 1987; Storkel, 2018). Thus, while phonologically complex, word-final targets (with or without morphological complexity) were found to be feasible, effective intervention targets for both participants, these findings also demonstrate a need for additional information about the effects of intervention targets with layers of complexity across both phonology and morphology.

Telepractice and the Complexity Approach

The transition to telepractice as a result of the COVID-19 pandemic provided opportunities to deliver fully remote assessment and intervention programs for children with SSD. Telepractice is a valid and appropriate service delivery model that can be used to deliver high-quality assessment and intervention to children with SSD (ASHA, n.d.). Importantly, SLPs across health care and academic settings have the ability to provide telepractice services, which help to address critical shortages in underserved or rural areas with limited access to clinical care providers (Dunkley et al., 2010). In addition, telepractice may be a necessary option for individuals with co-occurring disabilities and/or who experience socioeconomic barriers to accessing in-person services. However, gaps in telepractice research call for more evidence in support of the use of specific intervention practices (Hao et al., 2021).

With relevance to the current study, speech sound interventions delivered remotely have been shown to support children with SSD with similar effectiveness as compared to in-person services (Coufal et al., 2018; Grogan-Johnson et al., 2013). However, available studies have predominantly included normative target selection processes (Wales et al., 2017), rather than complex targets. Emerging evidence has demonstrated the appropriateness of a remote service delivery model for a complexity-based speech sound intervention: A pilot study of three participants receiving telepractice intervention with a complex cluster demonstrated that clinicians were able to adhere to the evidence-based elements of speech sound assessment, target selection, and intervention (John et al., 2023). Further confirmation of success in the telepractice context would extend what is known about complex target selection and support SLPs in considering service modality as they individualize approaches for their clients, thereby increasing positive outcomes for children with SSD.

Present Study

Complex targets are understood to be highly effective for children with SSD, though available studies largely feature word-initial targets. Notably, the word-final context in languages like English provides additional opportunities for adjusting the linguistic complexity of an intervention target via bound morphemes—though the extent to which this added layer of complexity is beneficial for children with SSD is not yet fully understood. Furthermore, investigating treatment effects in the context of telepractice provides much-needed evidence relevant to a service delivery model that increases access to services. To characterize the effects of complex word-final targets in speech sound intervention, the present study investigates whether children with SSD demonstrate (a) learning of complex word-final intervention targets, (b) generalization to untreated consonant singletons and clusters, and (c) increased overall phonological performance. It is hypothesized that all children will make gains following intervention, with greater gains observed when targets feature cross-domain linguistic complexity. Additionally, intervention effects are expected to differ across participants due to individual differences.

Method

The present study is part of a National Institutes of Health (NIH)–funded clinical trial (https://clinicaltrials.gov, No. NCT03977701) and was approved by the Institutional Review Board and Human Subjects Protection Program at San Diego State University (SDSU). This study employed a multiple-baseline, single-subject research design (Byiers et al., 2012; Gierut, 2008). All recruitment efforts and screening and assessment procedures were conducted remotely, in compliance with social distancing recommendations during the COVID-19 pandemic. Recruitment fliers were sent to child development and day care centers locally and across California. Information about the current study was also posted to Children Helping Science (https://childrenhelpingscience.com). In addition, school-based SLPs were asked to refer students. Caregivers were provided with a thorough review of the study protocol via phone or video call, and all participating caregivers provided electronically signed informed consent for their children to participate.

Participants

Eight English-speaking children with functional SSD, ages 3;3–6;9, were eligible based on the following criteria. All participants demonstrated a standard score at least 1.5 SDs below the mean on the Goldman-Fristoe Test of Articulation–Third Edition Sounds in Words subtest (Goldman, 2015) and had at least five consonants or consonant clusters missing from their sound system, based on performance on the Shorter Protocol for the Evaluation of English Phonotactics (Little PEEP; Barlow, 2012). Based on age and phonological profiles, all participants were candidates for complex intervention targets (e.g., Storkel, 2018). Participants were not enrolled in additional speech-language services, and due to the use of telepractice, families were required to have access to a laptop, tablet, or desktop computer with high-speed internet connection.

Additionally, all caregivers reported concerns with their child's speech sound production and/or intelligibility; no concerns regarding hearing, oral-motor abilities, fluency, or voice were noted by the caregivers or supervising clinicians. All participants demonstrated age-appropriate language abilities (Zimmerman et al., 2011; see Table 1), resided in a predominantly English-speaking household, presented with typical oral-motor skills on a remote oral mechanism examination, and successfully completed initial assessments via telepractice with minimal to moderate caregiver support.

Table 1.

Participant information pre-intervention.

ID	Age (years;months)	Sex	GFTA-3 SIW	PLS-5 TLS-SS	PLS-5 AC/EC	PCC (%)
Mia	4;10	F	56	107	107/107	55.83
Megan	4;9	F	76	125	116/130	76.86
Mabel	5;5	F	67	109	104/112	70.95
Monroe	4;4	M	75	115	113/114	70.05
Benjamin	3;3	M	72	125	126/120	46.37
Burton	4;5	M	68	91	96/88	64.21
Blossom	4;6	F	69	99	104/95	62.12
Briana	6;9	F	49	91	88/96	75.94

Note. GFTA-3 = Goldman-Fristoe Test of Articulation–Third Edition; SIW = Sounds in Words subtest (standard score; M = 100, SD = 15); PLS-5 = Preschool Language Scale–Fifth Edition; TLS = total language score; AC = Auditory Comprehension subtest; EC = Expressive Communication subtest (standard score; M = 100, SD = 15); PCC = percentage of consonants correct; F = female; M = male.

Procedures

All assessment and intervention procedures were conducted via telepractice. Sessions were held via Zoom video conferencing software that met compliance under the Health Insurance Portability and Accountability Act. All children participated in an initial assessment battery (Pre). Following Pre and prior to the start of intervention, three Baseline probes were administered to establish the stability of each child's speech sound abilities and identify an individualized treatment target (see below). Assessments to evaluate intervention effects were administered at the midpoint of intervention (Mid) and immediately following the intervention phase (Post), as well as at 2 weeks post-intervention (Follow-Up 1) and 2 months post-intervention (Follow-Up 2). Assessment materials and measures are described below.

One-on-one intervention sessions were provided by trained graduate student research assistants (session leads), with each participant consistently interacting with the same session lead; sessions were supervised by a licensed and credentialed SLP (either the first or the second author). Intervention sessions took place 3 times per week for 6 weeks (18 total sessions) and lasted 35–45 min. Intervention was provided in two phases: Sessions 1–9 facilitated production of the treatment target in imitation, and Sessions 10–18 facilitated spontaneous production (Gierut, 2015). The imitation phase consisted of primarily drill and drill-play activities (e.g., Shriberg & Kwiatkowski, 1982) and facilitated production of the target sound in isolation, at the syllable level, and within the treatment words (e.g., Gierut, 2015). As participants grew familiar with target words, practice opportunities were embedded into individualized activities that matched the interests of each child (e.g., digital sticker activities featuring preferred themes or characters). The spontaneous phase aimed to provide opportunities for children to independently produce their target sound within sentences, stories, and conversational contexts; for example, stories featuring individualized target verbs were created using slideshow software (e.g., a story about a soccer match to target kick/3s), and children were invited to narrate. Across both intervention phases, feedback was provided consistently, with a goal of feedback for each production attempt. Feedback included information about accuracy of the attempt, specific feedback (e.g., placement cues), recasts, and opportunities for children to self-evaluate, with feedback including verbal, visual, and gestural cues. For all sessions, clinicians facilitated maximal practice opportunities for the child to produce their target sound (mean productions per session = 107, SD = 16.05; Powell et al., 1998). Fidelity checklists, which characterized intervention activities, clinician feedback, and children's production attempts, were provided to session leads. Furthermore, a trained research assistant observed the session lead and completed the fidelity checklist during intervention sessions (available for 92% of sessions). Telepractice considerations (e.g., environmental setup, frequent collaboration with caregivers) were individualized to each child based on their interests, support needs, and resources (John et al., 2023).

Materials and Measures

All materials were created or modified specifically for use via telepractice, utilizing presentation software and screen sharing functions.

Assessment Materials

All participants completed a phonological probe (Little PEEP; Barlow, 2012) at Pre, Mid, Post, Follow-Up 1, and Follow-Up 2. The Little PEEP is a single-word speech sound probe designed to elicit multiple productions of all singletons and consonant clusters across word positions in the English language. For each item on the probe, children were shown a picture associated with the target word and provided with a question or phrasal cue (e.g., “What is this? It's a … ”), or with the target word and short delay as needed (e.g., “They go to the [playground]. Where do they go? To the … ”). Two different versions of the Little PEEP were rotated for each subsequent administration. Both versions included the same word list, with differing picture stimuli and item order, and order of presentation was counterbalanced across participants. Measures of primary interest (see the Measures and Analyses section, below) were derived from each child's performance on the Little PEEP across assessment sessions.

Data from the Little PEEP at Pre were used to derive Baseline probes, which were individualized to each child. Probes contained a subset of the Little PEEP items featuring those singletons and word-final consonant clusters that a child had produced with 0%–34% accuracy at the initial assessment (Combiths, Barlow, & Sanchez, 2019; e.g., if a child produced /s/ with < 34% accuracy, their baseline probe contained opportunities to produce /s/ across word positions). Baseline probe administration followed the same procedure as the Little PEEP, with alternating picture stimuli at subsequent assessment points. Performance on these probes established stability of speech sound production, allowing for selection of each child's individualized monitored sounds and intervention target.

Phonetic Transcription

Participants' single-word productions on the Little PEEP were transcribed by research assistants trained in phonetic transcription and the International Phonetic Alphabet. Transcriptions were entered in Phon software (Version 2.2; Rose & Hedlund, 2017) and analyzed to derive the outcomes reported in the current study. To calculate transcription reliability, 20% of each session was transcribed separately by research assistants who were unaware of the original transcription. Across all sessions, average interrater reliability was 84.5%.

Intervention Targets and Materials

All treatment targets were two-element consonant clusters in the word-final position and were stable at 0% accuracy at Pre and all three baselines (i.e., were unknown to the child). Target clusters were presented in the context of six treatment words, all of which were verbs. Four participants received monomorphemic targets (i.e., [−ks]; they mix) that were morphologically simple, and four received bimorphemic targets (i.e., [−ks]; she pick/3s) that were morphologically complex (see Table 2). Consistent with a single-subject design, participants were pseudorandomly assigned to a target condition (e.g., Gierut & Morrisette, 2015). Each participant received a cluster that was appropriate for their phonological system and that allowed for use of target verbs in intervention, with comparable numbers of children receiving morphologically simple and complex targets.

Table 2.

Individualized intervention targets and treatment words.

ID	Target	Morphological complexity	Treatment words
Mia	[−ks]	Monomorphemic	mix, max, fix, nix, wax, unbox
Megan	[−ɹd]		guard, board, reward, hoard, discard, record
Mabel	[−ɹd]		guard, board, reward, hoard, discard, record
Monroe	[−ɹs]		pierce, course, endorse, source, morse, parse
Benjamin	[−ɹz]	Bimorphemic	soars, hears, dares, tours, cares, repairs
Burton	[−ɹd]		soared, dared, toured, cared, repaired, veered
Blossom	[−ks]		packs, unpacks, picks, kicks, knocks, unlocks
Briana	[−ɹd]		soared, dared, toured, cared, repaired, veered

Note. All intervention targets were phonologically complex (i.e., were clusters). Clusters that did not feature a tense and agreement morpheme were morphologically simple (i.e., monomorphemic), and those that featured third-person singular or regular past tense were morphologically complex (i.e., bimorphemic).

At the beginning of each intervention session, a daily accuracy probe was administered to assess participants' production of their word-final target consonant cluster. Presentation of this probe was similar to that of the Little PEEP and consisted of each child's six treatment words repeated 3 times (18 total opportunities). Mirroring the two intervention phases, productions were facilitated in delayed imitation for Sessions 1–9 and independently for Sessions 10–18. In contrast to the remainder of the session, session leads did not provide feedback on target cluster production for the duration of this accuracy probe.

Measures and Analyses

To capture multifaceted changes in phonological knowledge as a result of intervention, we analyzed both learning of the target cluster and evidence of broader, systemwide generalization outcomes. First, intervention target accuracy was derived from the daily accuracy probe and reflected the treating clinician's observation. Omissions, substitutions, and other errors within the target cluster were characterized as inaccurate. The number of on-target productions was divided by the total opportunities (18) and multiplied by 100, yielding an accuracy percentage. Increasing accuracy reflected learning the individualized target cluster, which was unknown to each child before intervention (i.e., 0% accurate across Pre and Baseline assessments), and the criterion for mastery was > 80% across three consecutive sessions. Though learning of discrete speech sounds that are treated during intervention is a valid measure of improvement for children with SSD, the emphasis of the complexity approach is on systemwide generalization after intervention; thus, we emphasize measures reflecting performance on untreated sounds and untreated words.

Accuracy on untreated speech sounds and untreated words is a widely used approach for capturing generalization (e.g., Combiths, Escobedo, et al., 2022; Gierut et al., 2015; Potapova et al., 2023). Here, these monitored sounds of interest were word-final singletons and clusters that were consistently at 0% accuracy prior to intervention, as determined by performance on the Little PEEP and subsequent Baseline probes. Given the consistent low accuracy for these consonants and clusters prior to intervention, comparisons of performance from Pre to Post offer strongest evidence of generalization (Gierut, 2008); monitoring was continued at Follow-Ups 1 and 2. Accuracy for monitored sounds at each time point was determined using Phon (Rose & Hedlund, 2017). Additionally, effect size was calculated for each set of monitored sounds to determine the magnitude of change in phonological outcomes following intervention. Following Gierut (2015), effect size was calculated as standard mean difference (SMD). The numerator in the equation was formed by subtracting the average performance on monitored sounds pre-intervention (Pre, Baselines 1–3) from the average performance during and immediately post-intervention (Mid, Post). The denominator in the SMD equation represents variance pre-intervention. As the monitored singletons and clusters were consistently produced with 0% accuracy across all pre-intervention sessions, variance in these participants' performance could not be calculated and, per Gierut's (2015) recommendation, .02 was used as the denominator.

Beyond generalization, broader measures were used to investigate change in single-word productions across Pre, Post, and Follow-Up assessments. First, phonological mean length of utterance (PMLU; Ingram, 2002) was calculated using Phon (Rose & Hedlund, 2017). PMLU is the sum of the number of accurate consonants produced and the number of total segments produced; by crediting both accuracy and length, increasing PMLU reflects productions that increasingly approximate target words and contain increasingly complex syllable shapes. For example, a child may produce [da] for dogs /dɑgz/ at Pre and [dɑks] at Post; while accuracy has not increased in this example, PMLU increases from 3 at Pre to 5 at Post. Second, PMLU-Clusters captures localized change in cluster production (Potapova et al., 2023) and was calculated by hand. Like PMLU, these scores are the combined sum of accurate consonants and total segments produced, but the scope of focus is the cluster, rather than a whole word. In the previous example, [dɑ] would be scored as 0 for PMLU-Clusters word-finally and [dɑks] as 2 (0 correct consonants in the word-final cluster + 2 segments).

Results

Results are presented in alignment with our research aims, addressing learning of the intervention target, generalization to monitored sounds, and broader phonological change.

Learning of Intervention Target: Daily Accuracy

Each child was assigned an individualized, unknown word-final consonant cluster as their intervention target, and progress in learning the target was measured with a daily accuracy probe. Of the four children assigned monomorphemic clusters (e.g., [−ks] in mix; see Table 2), two demonstrated learning of their target clusters (see Figure 1A). Mia demonstrated fluctuations in accuracy, with 16 total accurate productions of her target, [−ks], between Sessions 13 and 18. Megan demonstrated a steadier positive trajectory, with 94 accurate productions of her target, [−ɹd], between Sessions 3 and 18. In contrast, Mabel, assigned [−ɹd], and Monroe, assigned [−ɹs], did not produce their intervention targets accurately during any daily accuracy probe.

Figure 1.

Learning of intervention targets. Accuracy for individualized intervention targets prior to intervention (Pre and Baselines [BLs] 1, 2, and 3) and during intervention (Sessions 1–18). (A) Accuracy for children assigned monomorphemic targets. (B) Accuracy for children assigned bimorphemic targets.

Variability in target learning was also observed for the four children whose targets featured bimorphemic consonant clusters (e.g., [−ks] in pick/3s; see Figure 1B). Blossom demonstrated 29 accurate productions of her target, [−ks], between Sessions 4 and 18. Her accuracy fluctuated, reaching 50% during Session 16. Benjamin, assigned [−ɹs], demonstrated fluctuating accuracy with few accurate productions: seven in total between Sessions 9 and 13. The two remaining participants, Burton and Briana, both assigned [−ɹd], did not demonstrate any accurate productions during any accuracy probes.

Overall, no child with either target type mastered their intervention targets (i.e., demonstrated > 80% accuracy across three sessions). Importantly, complex intervention targets were selected for their potential to facilitate systemwide phonological change. As such, we turn our attention to generalization to untreated sounds.

Generalization to Monitored Sounds

Monitored sounds were those word-final singleton consonants and word-final consonant clusters that were at 0% accuracy at Pre and across each of three Baseline sessions. Accordingly, monitored sounds were unique to each child (see Table 3) and were likely particularly challenging, as errors (e.g., omissions, substitutions) were observed on each opportunity to produce the sound or cluster across four assessment points; furthermore, these consonants and clusters were untreated and tested in untreated words. Gains in accuracy from Pre to Post thus provide strongest evidence of generalization (Gierut, 2008).

Table 3.

Monitored sounds and standard mean difference effect size.

Participant	Monitored sounds in word-final position		Baseline (%)	Intervention (%)	Effect size
Mia	Singletons	ʤ, s, ʃ, z, θ	0	10.53	5.26 (medium)
	Clusters	bz, fs, ks, lps, lts, lz, mps, mz, nz, ŋks, ɹf, ɹg, ɹm, ɹn, ɹs, ɹt, ɹts, ɹz, sk, st, ts, θs	0	6.41	3.21 (medium)
Megan	Singletons	ʤ, ʃ, ʧ	0	46.45	23.22 (large)
	Clusters	ɹd, ɹf, ɹg, ɹk, ɹm, ɹn, ɹs, ɹt, ɹts, ɹʧ	0	3.85	1.92 (small)
Mabel	Singletons	ɹ	0	0.00	—
	Clusters	ld, lk, lks, lts, ɹd, ɹf, ɹg, ɹk, ɹm, ɹn, ɹs, ɹt, ɹts, ɹʧ, ɹz	0	5.26	2.63 (medium)
Monroe	Singletons	ð, ʃ	0	43.75	21.88 (large)
	Clusters	lks, lts, ɹd, ɹf, ɹg, ɹk, ɹm, ɹn, ɹs, ɹt, ɹts, ɹʧ, ɹz, θs	0	0.00	—
Benjamin	Singletons	ð, ɡ	0	5.56	2.78 (medium)
	Clusters	ld, lks, ɹd, ɹf, ɹg, ɹk, ɹm, ɹn, ɹs, ɹt, ɹts, ɹʧ, ɹz, θs	0	8.33	4.17 (medium)
Burton	Singletons	ð, ʧ	0	0.00	—
	Clusters	lks, ɹd, ɹf, ɹg, ɹm, ɹn, ɹs, ɹt, ɹts, ɹʧ, ɹz, θs	0	3.57	1.79 (small)
Blossom	Singletons	ð, ɡ, ɹ	0	3.03	1.52 (small)
	Clusters	ks, ld, lk, lks, lts, ɹd, ɹf, ɹg, ɹk, ɹm, ɹn, ɹs, ɹt, ɹts, ɹz, sk, st, θs	0	11.13	5.56 (medium)
Briana	Singletons	ð, ɹ	0	0.00	—
	Clusters	ɹd, ɹf, ɹg, ɹk, ɹm, ɹn, ɹs, ɹt, ɹts, ɹʧ, ɹz	0	0.00	—

Note. Baseline reflects performance at Pre and across Baselines, and Intervention reflects performance at Mid and Post. Effect size was not calculated (—) when no change in performance was observed at Mid or Post.

Monitored Word-Final Singletons

Of the four children with monomorphemic intervention targets, three—Mia, Megan, and Monroe—demonstrated increased accuracy for monitored word-final singletons immediately following intervention, with average accuracy rates ranging from 17.54% to 62.50% at Post (see Figure 2A). Gains increased at Follow-Up assessments for Mia and Megan, reaching as high as 95.00% for Megan, and Monroe maintained an increase relative to Pre at Follow-Up 2. The remaining participant, Mabel, demonstrated no change in accuracy from Pre to Post intervention, demonstrating 0% accuracy at each assessment point. Of note, this participant only had one monitored word-final singleton, [ɹ]; all other participants had at least two monitored singletons.

Figure 2.

Performance on monitored word-final sounds at pre-intervention (Pre and Baselines [BLs] 1, 2, and 3) and during/following intervention (Mid, Post, Follow-Up 1, and Follow-Up 2). (A) Monitored word-final singletons for participants assigned a monomorphemic intervention target. (B) Monitored word-final singletons for participants assigned a bimorphemic intervention target. (C) Monitored word-final clusters for participants assigned a monomorphemic intervention target. (D) Monitored word-final clusters for participants assigned a bimorphemic intervention target.

For children with bimorphemic targets, only one child demonstrated increased accuracy for monitored singletons immediately at Post (see Figure 2B): Blossom produced her monitored singletons with 6.06% accuracy immediately following intervention, and her gains increased at subsequent assessment points. Two additional children demonstrated increased accuracy at other points during or following intervention: Benjamin demonstrated transient gains, with accuracy rates of 11.11% at the mid-point of intervention and again at Follow-Up 1, and Burton demonstrated steadily increasing accuracy on his monitored word-final singletons starting 2 weeks after intervention concluded, reaching 28.57% at Follow-Up 1 and 57.14% at Follow-Up 2. Finally, Briana demonstrated no change in monitored word-final singletons following intervention, producing this set with 0% accuracy at all assessment points.

Monitored Word-Final Clusters

Accuracy on monitored word-final clusters, which share both word position and structure with the intervention target, was also of interest. From stable, 0% accuracy across multiple pre-intervention assessments, three of four children with monomorphemic targets—Mia, Megan, and Mabel—demonstrated increased accuracy at Post, ranging from 5.26% to 12.82% (see Figure 2C). Megan and Mabel increased in accuracy at subsequent time points, and Mia decreased in accuracy at Follow-Up 1 before demonstrating a particularly high accuracy rate, 56.76%, at Follow-Up 2. Monroe maintained 0% accuracy immediately following intervention and demonstrated transient gains thereafter, with an accuracy rate of 5.56% at Follow-Up 1.

Two children with bimorphemic targets demonstrated gains on monitored word-final clusters at Post (see Figure 2D). Burton produced clusters in his monitored set with 7.14% accuracy, and Blossom produced clusters in her set with 11.54% accuracy; only Blossom maintained these gains at subsequent assessments. Benjamin again demonstrated transient gains, producing monitored word-final clusters with 16.67% accuracy at Mid and with 5.56% accuracy at Follow-Up 2, and Briana maintained 0% accuracy at all assessment points.

Effect Size

The magnitude of observed intervention effects was calculated as effect size, following Gierut (2015). Starting with monitored singletons (see Table 3), effect sizes were generally greater for children with morphologically simple targets. Mia demonstrated a medium effect size (SMD = 5.26), and Megan and Monroe each demonstrated large effect sizes (SMD = 23.22 and 21.88, respectively). Of the children with morphologically complex targets, Benjamin demonstrated a medium effect size (SMD = 2.78) and Blossom demonstrated a small effect size (SMD = 1.52). As Burton did not demonstrate increases in accuracy on monitored singletons until Follow-Up 1, his effect size was not calculated; similarly, an additional child with each target type—Mabel and Briana—did not improve in their production of monitored word-final singletons at any time point, and thus, effect size was not calculated.

Effect sizes were also calculated for monitored word-final clusters (see Table 3). Of children with monomorphemic targets, Mia and Mabel demonstrated medium effect sizes (SMD = 3.21 and 2.63, respectively) and Megan demonstrated a small effect size (SMD = 1.92). Monroe did not demonstrate an increase in accuracy on monitored clusters until 2 weeks after intervention, and as such, effect size was not calculated. Of children with bimorphemic targets, Benjamin and Blossom demonstrated medium effect sizes (SMD = 4.17 and 5.56, respectively), and Burton demonstrated a small effect size (SMD = 1.79). As noted above, no change was observed in word-final cluster production for Briana at any time point.

Taking monitored word-final singletons and clusters together, six of eight participants, including all participants with monomorphemic targets, demonstrated generalization to untreated sounds and clusters in untreated words immediately following intervention (i.e., at Post), with most continuing upward trends at Follow-Ups; a seventh participant demonstrated gains at time points other than Post. Overall, gains demonstrated by children with monomorphemic targets were more robust and more stable at subsequent assessment points.

Additional Evidence of Phonological Change

PMLU and PMLU-Clusters

PMLU adds depth to accuracy measures, as it credits productions for both accuracy and length. Here, we include both PMLU, a whole-word measure, and PMLU-Clusters, measured in word-final position (Potapova et al., 2023). For participants with monomorphemic targets, PMLU prior to intervention ranged from 6.18 to 7.28. Three children—Mia, Megan, and Mabel—demonstrated gains in PMLU following intervention, with increases of 0.52, 0.11, and 0.21, respectively (see Figure 3A), at Post. For each of these children, gains were sustained and at least minimally increased at Follow-Up sessions. Monroe's PMLU remained relatively stable from 6.75 at Pre to 6.73 at Post, reaching 6.78 at Follow-Up 2.

Figure 3.

Phonological mean length of utterance (PMLU) and PMLU-Clusters prior to intervention (Pre) and following intervention (Post, Follow-Up 1, and Follow-Up 2). (A) PMLU for participants assigned a monomorphemic intervention target. (B) PMLU for participants assigned a bimorphemic intervention target. (C) PMLU-Clusters for participants assigned a monomorphemic intervention target. (D) PMLU-Clusters for participants assigned a bimorphemic intervention target.

For children with bimorphemic cluster targets, increases in whole-word PMLU from pre- to post-intervention were also observed. Pre-intervention, PMLU scores ranged from 5.69 to 7.13. Following intervention, three of four participants—Benjamin, Blossom, and Briana—demonstrated gains of 0.17, 0.20, and 0.30, respectively, at Post (see Figure 3B). Benjamin and Blossom continued in an upward trajectory at later assessment points, and Briana sustained gains relative to Pre. In contrast, Burton's productions decreased in PMLU from 6.47 at Pre to 6.28 at Post, though this decrease was temporary: Productions at Follow-Ups 1 and 2 reflected increased length and accuracy relative to Pre, with PMLU scores of 6.77 and 6.83, respectively.

PMLU-Clusters similarly awards credit for productions of increasing length and accuracy; however, rather than whole-word comparisons, this measure isolates cluster productions. For children with morphologically simple targets, PMLU-Clusters in word-final position ranged from 2.44 to 3.72 at Pre. Most notably, Mia and Mabel demonstrated increases of 0.38 and 0.05, respectively, at Post, with continued upward trajectories by the final assessment point (see Figure 3C). In contrast, Megan and Monroe each demonstrated decreased PMLU-Clusters immediately after intervention and slight upward trends at Follow-Ups.

For children with morphologically complex targets, scores for PMLU-Clusters in word-final position ranged from 2.65 to 3.52 at Pre. Three children—Benjamin, Blossom, and Briana—demonstrated productions of increasing length and accuracy following intervention (see Figure 3D), with increases of 0.28, 0.17, and 0.09, respectively. These gains were sustained at subsequent assessment points relative to performance at Pre. Burton's PMLU-Clusters decreased from 3.52 at Pre to 3.33 at Post and remained in that range at both Follow-Ups.

Altogether, six of eight children produced single words with increasing accuracy and length immediately following intervention (i.e., at Post), as captured by whole-word PMLU and PMLU-Clusters in word-final position. Both of the remaining children—Monroe and Burton—demonstrated gains at later time points (in PMLU-Clusters and PMLU, respectively).

Results Summary

Table 4 represents trends across outcome measures. Recall that learning of the individualized target represents a narrow measure of intervention effects (i.e., change on treated target in treated words), whereas generalization and PMLU measures capture broader change. To simplify comparisons across measures, Table 4 includes visual representations of trends. In brief, all children made gains across multiple measures following intervention, with more consistent evidence of growth for children with monomorphemic targets. Within each cluster type (i.e., mono-, bimorphemic), individual trends were observed, with some children demonstrating more consistent evidence of growth (e.g., monomorphemic target: Mia, bimorphemic target: Blossom) than others (e.g., monomorphemic target: Monroe; bimorphemic target: Briana).

Table 4.

Intervention effects summary.

	Learning of target across intervention	Monit. WF singletons: effect size	Monit. WF clusters: effect size	PMLU at Post and Follow-ups	PMLU-Clusters at Post and Follow-ups
Mia	↗	↗	↗	↗↗	↗↗
Megan	↗	↗	↗	↗↗	↘→
Mabel	→	→	↗	↗↗	↗↗
Monroe	→	↗	→	→→	↘↗
Benjamin	↗	↗	↗	↗↗	↗↗
Burton	→	→	↗	↘↗	↘→
Blossom	↗	↗	↗	↗↗	↗↗
Briana	→	→	→	↗↗	↗↗

Note. Phonological change is summarized as follows: ↗ indicates gains;↘ indicates decreased performance; and → represents stable performance. Learning of target = accuracy on daily probes. Any correct productions during any of 18 intervention sessions are interpreted as gains. Monit. WF singletons = monitored word-final singletons. Evidence of small, medium, or large effect sizes across post and follow-ups are interpreted as gains. Monit. WF clusters = monitored word-final clusters. Evidence of small, medium, or large effect sizes across Post and Follow-ups are interpreted as gains. PMLU = phonological mean length of utterances. Changes from Pre to Post are represented by the first arrow; changes from Pre to Follow-ups are represented by the second arrow. Increases in performance relative to pre are interpreted as gains. PMLU-Clusters = phonological mean length of utterances–Clusters, calculated in word-final position. Changes from Pre to Post are represented by the first arrow; changes from Pre to Follow-ups are represented by the second arrow. Increases in performance relative to Pre are interpreted as gains.

Discussion

The complexity approach in speech sound intervention has a particularly rich evidence base, guiding the selection of individualized treatment targets for children with SSD (Storkel, 2018). Critically, much of this work has investigated the effects of word-initial complex targets (e.g., [pl−]). The current study extends promising case studies featuring word-final consonant clusters (e.g., [−ks]; Potapova et al., 2023; see also Combiths, Barlow, Richard, & Pruitt-Lord, 2019). For children with SSD, word-final phonological processes (errors) are not uncommon (Weismer et al., 1981), and these omissions and substitutions may impact intelligibility and the expression of grammatical markers in languages with suffixes. Thus, the current study is a meaningful addition to how complex targets may be utilized in clinical practice.

Here, we investigate intervention effects following treatment of a word-final consonant cluster. Eight children (3;3–6;9) completed 18 sessions targeting an individualized word-final consonant cluster that was consistently produced with 0% accuracy prior to intervention. Each target was phonologically complex (i.e., a cluster) and presented in the context of six verbs. Targets were either morphologically simple, such as [−ks] in they mix, or morphologically complex, featuring a tense and agreement marker, such as [−ks] in she pick/3s (see Table 2). Thus, we explore the unique potential afforded by word-final position—the manipulation of morphological complexity over and above the phonological complexity of clusters relative to singletons. An important feature of this study is that all assessment and intervention sessions were conducted via telepractice. This service delivery model offers substantial benefits, including increasing clients' access to services (Fairweather et al., 2016; Molini-Avejonas et al., 2015) and has been increasingly implemented in the wake of COVID-19 (Bolden, 2022; Kollia & Tsiamtsiouris, 2021). The present study integrates a theoretically informed intervention (i.e., the complexity approach) with this valuable service delivery model (John et al., 2023).

Trends in Intervention Effects

Following intervention, all participants in the present study demonstrated positive change across multiple measures of speech sound production. Recall that complexity across language domains has been found to facilitate generalization following intervention (Kiran, 2008; Owen Van Horne et al., 2018; Thompson & Shapiro, 2007) and complexity arises from multiple sources concurrently (Storkel, 2018). In the present study, all clusters were phonologically complex (i.e., were clusters), and linguistic sources of complexity were further manipulated by implementing both morphologically simple and complex clusters. Furthermore, child-internal factors–including pre-intervention phonological knowledge—are understood to contribute to complexity (Gierut & Champion, 2001; Gierut et al., 1987). Below, we explore trends in our findings with respect to both linguistic and child-internal sources of complexity.

Linguistic Complexity and Intervention Effects

With four participants receiving each type of intervention target, comparisons between effects of mono- and bimorphemic targets must be made with caution. Nevertheless, present trends add to our understanding of how complexity may be leveraged in intervention and motivate future research. Overall, counter to initial hypotheses, there was more robust evidence of intervention effects for children whose targets were morphologically simple, all of whom demonstrated generalization (i.e., improvement on monitored singletons and/or clusters) immediately at Post, compared to two of four children with bimorphemic targets. Effect sizes, which facilitate comparison across contexts and studies (Gierut, 2015), revealed a similar pattern for monitored singletons: Intervention effects were generally larger for children with morphologically simple targets. In contrast, effect sizes for monitored clusters were comparable for children with both target types, and all children demonstrated phonological growth across multiple measures. Thus, unlike patterns observed by Potapova et al. (2023), in which the child who received intervention on a bimorphemic cluster demonstrated greater gains than the child with a monomorphemic cluster, present findings do not indicate that maximizing complexity across linguistic domains is superior in facilitating generalization or phonological growth. Importantly, children in the earlier case studies differed from one another in phonological knowledge prior to intervention, as did participants in the current work. Thus, findings across studies highlight the importance of considering child-internal sources of complexity alongside linguistic sources of complexity, a topic we return to below.

With respect to linguistic complexity, we may also consider constituent elements in each cluster, which may contribute to tiers of phonological complexity (Gierut, 1999). Consonants were largely paired across target types in the present study, with mono- and bimorphemic forms of [−ks] and [−ɹd] and monomorphemic [−ɹs] bearing similarity to bimorphemic [−ɹz] (see Table 2). Within each level of morphological complexity, the most widespread gains were demonstrated by Mia and Blossom, who were each assigned [−ks]. Potentially, this combination of sounds was particularly effective in stimulating change. The remaining targets featured [ɹ] and an additional consonant, and children with these targets performed more variably. The phoneme [ɹ] is notable: This late-developing sound is often targeted in speech sound intervention (e.g., see Krueger & Storkel, 2023), and [ɹ] production patterns may be particularly resistant to change (Preston et al., 2020). Potentially, these characteristics impacted the complexity of the targets such that [ɹ] + consonant targets do not optimize generalization (Morisette et al., 2006). However, Megan, assigned [−ɹd], made the greatest gains in daily accuracy and demonstrated positive change across most measures, and all children with [ɹ] + consonant targets demonstrated some evidence of phonological growth. Thus, our findings suggest that a variety of word-final clusters may serve to facilitate phonological growth, and future work is needed to specify target selection that optimizes gains.

Individual Differences and Intervention Effects

As noted above, relative complexity of targets also arises from child-internal factors, such as their linguistic knowledge (Gierut et al., 1987). Indeed, generalization following intervention may depend on a sufficient degree of phonological knowledge (Gierut & Champion, 2001). It is possible that, analogously, a child's knowledge of morphosyntax would bear on the relative complexity of morphologically complex targets. Of the children assigned bimorphemic targets, the two children to demonstrate greatest change following intervention (Blossom, Benjamin) were the children with relatively higher overall language performance on a standardized assessment (see Table 2), and though all participants' language performance was age-appropriate, children assigned bimorphemic targets generally demonstrated lower total language scores than children with relatively simple targets. Potentially, linguistic knowledge contributed to trends in intervention effects for all participants. In future work, it would be valuable to investigate the interaction of morphosyntactic knowledge, including measures beyond standard scores, and target complexity to identify impacts on generalization.

Beyond morphological knowledge, other individual differences may be relevant to outcomes. Nevertheless, intervention effects were observed for children with relatively less phonological knowledge (e.g., Mia, with relatively low PMLU at Pre and relatively many monitored sounds) and for children with relatively advanced phonological systems (e.g., Megan, with relatively high PMLU and PMLU-Clusters at Pre and relatively few monitored sounds). Gains were also observed for children who did and did not demonstrate learning of their targets (e.g., Blossom and Monroe), as well as for children that were relatively younger and older (e.g., Benjamin, 3;3, and Briana, 6;9). Continued work with larger samples may clarify how these and other individual characteristics may be considered as targets are selected. Presently, findings suggest that treatment of an unknown, word-final consonant cluster may facilitate phonological growth for a range of English-speaking children with SSD in a relatively short period of time.

Limitations

While providing evidence from a larger sample of children with SSD than previous related work (Combiths, Barlow, Richard, & Pruitt-Lord, 2019; Potapova et al., 2023), the current study does not allow for broad generalization of findings, in part because all participants were English speakers. As word-final complexity (in phonology and morphology) is present in many languages, word-final clusters may be feasible targets across languages, and continued work is necessary to explore this potential cross-linguistically (Combiths, Escobedo, et al., 2022; see also Barlow, 2005). Another characteristic of the present sample is that participants demonstrated word-final errors that largely impacted cluster production, and relatively few singleton consonants were missing from pre-intervention repertoires. For children with more restricted inventories, other generalization patterns may be observed. Present participants also had the support and resources to participate in individual intervention sessions via telepractice 3 times per week. Using evidence-based practice, clinicians may integrate these findings with their judgment and their clients' perspectives to implement complex word-final targets in a manner appropriate for their local context (e.g., in person, group settings, differing session lengths; John et al., 2023). Future work may also compare outcomes following intervention of clusters to other target types presented in the relatively novel word-final position.

Summary and Clinical Implications

Following intervention targeting unknown word-final clusters, eight of eight English-speaking children demonstrated change across multiple measures of phonological performance. More stringently, six participants demonstrated generalization to untreated sounds and untreated words immediately following intervention. Of note, these gains were observed following a relatively short course of the intervention (18 sessions in 6 weeks), and learning of the target was not required for this broad growth to occur. Altogether, findings demonstrate that complex, word-final targets facilitated phonological growth for English-speaking children with SSD. When comparing progress across measures, patterns often converged; for example, a child may have shown increased accuracy on monitored sounds and an increase in PMLU from Pre to Post. In other cases, one measure (or type of measure) may have revealed progress not evident in the other (see Table 4). This adds to our understanding that assessment and progress monitoring are served by considering multiple measures and that measures of growth ought to reflect specific intervention goals. Furthermore, we note that increasingly descriptive measures that reflect graded improvement (e.g., PMLU, PMLU-Clusters), as compared to binary accuracy judgments, serve an important purpose (see also Combiths, Pruitt-Lord, et al., 2022). Importantly, intervention effects were observed following sessions conducted entirely via telepractice. As such, present findings further affirm the use of this important service delivery model that increases access to services, including for underserved communities.

Targets included in the present intervention reflected complexity from multiple sources: child-internal factors (i.e., clusters were unknown pre-intervention) and linguistic features (i.e., consonant clusters, with and without morpheme boundaries). Given these multiple, interacting layers of complexity, children (3;3–6;9) with SSD successfully participated in sessions, indicating that complex targets are not only effective but also feasible. As we increase our understanding of complex targets, we are increasingly able to optimize targets to serve client needs, and targets featuring cross-domain complexity may have the additional benefit of serving goals related both to speech sounds and grammar. Thus, these targets may be beneficial for children with concurrent difficulties in speech sound production and grammar (Combiths, Barlow, Richard, & Pruitt-Lord, 2019) and be functional in group intervention settings where children have differing goals. In summary, present findings on word-final complex clusters targeted via telepractice add to an evidence base that clinicians may draw from—alongside their own judgment and client perspectives—to individualize intervention to serve client needs.

Data Availability Statement

The data sets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Funding Statement

This work was supported by National Institutes of Health Grant NIDCD R21 DC017201 awarded to Jessica Barlow and Sonja Pruitt-Lord, and by the San Diego State University College of Health and Human Services COVID-Affected Research Enterprise Stimulus Award.