Generative Language Intervention for Young Children With Down Syndrome Using Augmentative and Alternative Communication: A Randomized Controlled Trial

Abstract

Purpose:

Children with Down syndrome often have poor speech intelligibility, which can mask expressive language competence; this, in turn, can lead to serious misconceptions about overall competence and intellectual abilities. Although aided augmentative and alternative communication (AAC) can be used to bridge these gaps, children with Down syndrome are not always provided with consistent access to focused AAC language intervention supports. The primary goal of this study was to evaluate the effect of implementing the AAC Generative Language Intervention (AAC-GLI) approach on the aided expressive grammar productions of young children with Down syndrome.

Method:

A randomized controlled trial was used to evaluate the results. The study included a control group and an intervention group, with the families of both groups participating in half-day AAC implementation workshops and all children receiving AAC devices to use throughout the course of the study. The intervention group also received 4 months of twice-weekly play-based AAC-GLI intervention sessions. Progress was measured using a mean length of utterance (MLU) specially designed for aided communicators (weighted MLU in symbols [W-MLUSym]).

Results:

Strong effects indicated superior performance on W-MLUSym for the intervention group, despite reduced enrollment and increased attrition yielded by the COVID-19 pandemic.

Conclusions:

AAC-GLI can be used to teach young children with Down syndrome to improve their aided expressive grammar skills. Providing AAC language intervention for young children with Down syndrome can be a critical step to support ongoing expressive language development and use as well as overall functional communication.

For children with Down syndrome, who account for approximately one out of 640 live births in the United States each year (Centers for Disease Control and Prevention, 2024), speech intelligibility is a significant issue. Parents report that their preschoolers with Down syndrome have difficulty being understood an average of 67% of the time, with this challenge reportedly remaining over 50% throughout the life span (Kumin, 1994). Direct measures indicate long-term issues with intelligibility, with the average adolescent with Down syndrome being approximately 50% intelligible, a moderate-to-severe impairment (Landa et al., 2014). Critically, cognitive abilities in Down syndrome often exceed speech abilities (Cleland et al., 2010); that is, unintelligible speech often masks underlying intellectual and communication capabilities for people with Down syndrome. These facts, combined with relatively strong receptive language skills (Martin et al., 2009), make augmentative and alternative communication (AAC) interventions excellent complements to spoken language interventions for children with Down syndrome to override challenges with speech. In fact, recent research findings demonstrate that implementing AAC interventions with children who have Down syndrome yields greater gains in both expressive vocabulary and intelligible spoken language than implementing spoken language interventions only with children who have Down syndrome (Romski et al., 2023).

Relatedly, children with Down syndrome are known to have expressive syntax issues, with syntax lagging behind nonverbal cognition (Abbeduto et al., 2007). By age 3;6 (years;months), only approximately half of children with Down syndrome are combining words (Berglund et al., 2001), and preschoolers with Down syndrome are far more likely to omit words in their utterances than their peers with language disorder (Caselli et al., 2008). Furthermore, grammatical morphology has been found to be particularly problematic (Abbeduto et al., 2007). Although some of these issues may be rooted in challenges with receptive grammar (Abbeduto et al., 2007), it is also possible that poor motor speech contributes to the limited output of preschoolers with Down syndrome. Using aided AAC bypasses speech issues, thereby providing children with Down syndrome with a communication mode that may allow them to use longer, more complex utterances (as well as more complex vocabulary) at earlier ages.

Yet, the corpus of research on the use of aided AAC with children with Down syndrome remains relatively sparse, with most studies focusing on developing pragmatic or early vocabulary skills (Barbosa et al., 2018); only limited attention has been paid to interventions supporting symbol combinations and more advanced expressive language. For example, Quinn et al. (2020) examined the effect of using an aided modeling-based cuing hierarchy (Binger et al., 2010; Kent-Walsh, Binger, & Hasham, 2010) on the symbol identification, rate of communication, and number of different words for four children with Down syndrome ages 3–5 years, with some participants also exhibiting symbol combinations in their productions. Furthermore, several children with Down syndrome have exhibited increases in productions of aided symbol combinations in response to direct child-focused and indirect communication partner–focused AAC interventions (Binger & Light, 2007; Kent-Walsh, Binger, & Malani, 2010). Despite these encouraging findings, two distinct gaps in the literature remain: a limited focus on teaching rule-based word combinations to children who use AAC (i.e., focus on aided language that follows the rules of spoken language) and an even more limited focus on teaching these skills to children with intellectual and developmental disabilities (IDDs)—including those with Down syndrome.

With some notable exceptions (Nigam et al., 2006; Tönsing et al., 2014), most studies aiming to increase children's aided AAC symbol combinations have focused on children with relatively intact receptive language skills (Binger et al., 2008, 2010) For example, two single-case experimental designs were implemented with preschoolers who used AAC to teach them to produce four different semantic relations (attribute–entity, agent–action–object, entity–locative, and possessor–entity; Binger, Kent-Walsh, King, & Mansfield, 2017; Binger, Kent-Walsh, King, Webb, & Buenviaje, 2017). Participants demonstrated significant gains in their aided language productions on most targets. To date, to our knowledge, no larger group studies such as randomized controlled trials (RCTs) have been conducted to examine the aided language grammatical skills of children with IDDs, including those with Down syndrome. Poor intelligibility for these children, combined with reports of only approximately half of them combining spoken words by age of 3.5 years (Berglund et al., 2001), makes this lack of research on aided grammatical development for children with Down syndrome concerning—particularly when developmental models clearly indicate the need to move beyond a focus on early developing pragmatic and semantic skills to also include a focus on building grammar skills (Binger et al., 2024). Overall, children with Down syndrome are prime candidates for producing rule-based utterances via AAC to complement their ongoing spoken language development.

To address this need, we developed the AAC Generative Language Intervention (AAC-GLI) program (Binger, Kent-Walsh, King, & Mansfield, 2017; Harrington et al., 2023). Specifically, AAC-GLI was designed to support the grammatical and semantic development of children who use AAC, including those with IDDs. AAC-GLI has three primary components: instructional techniques, technology implementation, and contexts for communication (see Figure 1). The instructional techniques include a range of techniques designed to support aided expressive language; Appendix A includes a detailed listing with described and linked video examples. These techniques include a range of elicitation and response techniques, such as wait time, aided models, and spoken models. Whenever possible, intervention techniques—such as binary choices, expansions, and extensions—are implemented to include both spoken and aided input. More specifically, an overarching technique is the use of grammatically complete spoken models (“Blue Hippo is in the airplane”), followed by an aided model that, on average, is 0.5–2 symbols longer than the child's typical aided utterance (BLUE HIPPO IN AIRPLANE). Aided models have long been shown to promote children's aided productions (Biggs et al., 2018). Additionally, children are encouraged to produce their messages using aided output. Past research indicates that such use of both augmented input and output support expressive language development (Romski et al., 2010).

Figure 1.

Participant flowchart. mo. = months.

The technology implementation component of AAC-GLI involves the use of speech-generating devices, including mobile devices such as iPads that are fitted with a widely available AAC app. Any number of AAC apps can be used, at least during the initial phases of the intervention. Initial displays use single-meaning graphic symbols (e.g., separate symbols for DOG, IS, IN, THE, and CAR), are context-specific (e.g., one display for a vehicles play routine and another for a baking play routine), contain diverse vocabulary with multiple parts of speech (nouns, verbs, adjectives, etc.), include early grammatical markers (e.g., progressive –ing), and do not require navigation to different displays during any given routine (see Figure 1). Thus, the child is taught to build rule-based phrases and sentences one word at a time. We view the use of these activity-specific displays, with all words needed for a particular activity on one display, as critical for the initial stages of grammar development. This minimizes cognitive demands, as children do not need to use their cognitive resources to search for vocabulary across multiple displays. In our preliminary studies, children quickly expanded utterance length and accuracy using this approach (Binger et al., 2008, 2010, 2011; Binger, Kent-Walsh, King, & Mansfield, 2017; Binger & Light, 2007; Kent-Walsh et al., 2015; Kent-Walsh, Binger, & Malani, 2010). In contrast, even typically developing preschoolers have difficulty locating words when they must search through multiple displays to locate words (Drager et al., 2003). We therefore view activity-specific displays as one critical component of a child's aided communication repertoire, although certainly not the only one (Binger et al., 2024).

The last component of AAC-GLI focuses on the contexts of intervention. Studies designed to promote the spoken and aided communication skills of children with developmental disabilities have stressed that intervention contexts must engage children and create the need to communicate (Kaiser & Roberts, 2013; Romski et al., 2010). For the current investigation, which focuses on children ages 3;0 to 5;11 with Down syndrome, age-appropriate symbolic play routines were used as the context for communication (e.g., vehicle and baking play routines). These types of contexts have been used successfully in previous research to promote communication with children at similar developmental levels (Kaiser & Roberts, 2013).

In summary, young children with Down syndrome are excellent candidates for aided AAC given their poor intelligibility and discrepancies between intelligibility and intellectual abilities. Yet, aided interventions for this population are still in their infancy, with a particular dearth of information available regarding AAC interventions designed to increase utterance length and complexity. Therefore, the specific research question for this study was: For preschoolers with Down syndrome, does implementing AAC-GLI with participating children, in addition to a 4-hr AAC caregiver/family member workshop, result in improved child aided utterance length compared with the child aided utterance length for a control group receiving only the workshop? Our hypothesis was that the AAC-GLI group would have superior outcomes compared with the control group on weighted mean length of utterance in symbols (W-MLUSym), a measure adapted from the traditional spoken language mean length of utterance in morphemes (MLUm) that adjusts for unique aided language issues (Lee et al., 2022).

Method

This study was an RCT (ClinicalTrials.Gov NCT03538925). Therefore, the CONSORT guidelines for reporting randomized trials (Tetzlaff et al., 2012) were followed to detail the method for this study, as depicted in Figure 1.

Participant Selection

Recruitment

Participants were recruited across both sites via flyers, social media posts, and e-mail communication with local Down syndrome organizations, community health care providers, and schools. Thirty-six participants passed an initial screening and were assessed for eligibility, with 29 successfully completing the baseline assessment (see Figure 1).

Inclusionary Criteria and Participant Profile Information

Table 1 contains participant details. Participants met the following inclusionary criteria: (a) were between ages 3;0 and 5;11 at the start of the study; (b) had a severe speech impairment, as defined by less than 50% intelligible with context on the Index of Augmented Speech Comprehensibility in Children (I-ASCC; Dowden, 1997); (c) spoke English as a primary language; (d) had a multimodal expressive vocabulary of at least 25 words on the MacArthur Communicative Development Inventories (CDI; Fenson et al., 2007) via any communication mode; (e) had at least one verb and one adjective selected on the CDI; (f) identified at least four different graphic symbols on the iPad app across at least three parts of speech (noun, verb, adjective, etc.); (g) had vision and hearing within normal limits or corrected to be within normal limits, as reported by their parents; (h) were able to select picture symbols on an iPad; and (i) had no diagnosis of autism. Twenty-eight of the 29 participants were reported by their parents to be enrolled in community preschool programs, and all but one participant was reported to have participated in speech-language therapy during the 12 months prior to enrolling in the study. Parents reported that their children received an average of 52 min of speech-language therapy per week across both sites.

Table 1.

Participant characteristics.

Child (I, C, MI)	Last Mx session	CA (years;months) Sex	Race/ethnicity	Primary (secondary) language	SES	Prior AAC use in monthsa	I-ASCC (% correct)	CDIb	TACL raw score sum (%ile)	PPVT SS	VABS Rec. v scale score (M = 15, SD = 3); AE in months	VABS Exp. v scale score (M = 15, SD = 3); AE in months	Symbol Axc (% correct)	Comprehension of semantic relationsc (% correct)
														AAO	Pos	Att	Loc
1 (I)	Maint.	5;8 M	A/NH	E	$75K–$99.9K	0	19	260	33 (< 1)	32	9 25	2 23	83	75	90	90	20
2 (C)	Maint.	4;8 F	W/NH	E	$100K–$149.9K	0	29	248	43 (8)	37	11 28	8 28	92	70	80	90	40
3 (I)	Maint.	4;2 M	W/NH	E	$150K–$199.9K	0	10	78	45 (14)	< 30	11 29	2 19	75	75	60	50	10
4 (C)	Maint.	3;8 M	O/H	E (S)	$25K–$34.9K	0	23	191	31 (10)	31	12 31	8 26	83	55	30	30	30
5 (I)	Month 1	3;4 M	W/H	E	$75K–$99.9K	0	0	74	23 (10)	< 30	12 26	4 21	75	40	30	50	20
6 (C)	Month 4	3;6 F	W/NH	E	NR	0	0	137	10 (3)	< 30	9 19	7 22	75	55	60	10	20
7 (C)	Month 3	4;7 F	W/ NH	E	$75K–$99.9K	6	35	345	24 (1)	< 30	11 32	7 26	100	55	80	50	20
8 (I)	Month 1	5;9 M	W/H	E (S)	$75K–$99.9K	0	48	347	30 (< 1)	34	8 22	5 27	92	50	100	80	50
9 (C)	Maint.	4;4 M	W/NH	E (ASL)	$75K–$99.9K	0	0	112	20 (1)	65	10 23	2 19	66	50	90	20	0
10 (I)	BL	4;11 F	B/NH	E	$100K–$149.9K	0	23	314	67 (21)	34	9 24	8 30	92	60	90	70	50
11 (I)	BL	3;8 M	W/NH	E	$100K–$149.9K	0	9	55	10 (2)	65	6 11	5 19	75	20	30	30	20
12 (I)	Month 3	4;6 M	W/H	E	$35K–$49.9K	0	3	115	48 (14)	71	9 22	4 21	75	35	70	40	40
13 (MI)	Maint.	4;2 M	W/NH	E	$150K–$199.9K	0	3	118	60 (30)	78	11 26	6 23	75	85	100	100	100
14 (MI)	Maint.	3;8 M	W/H	E (S)	NR	0	10	65	36 (18)	76	10 22	4 20	50	50	30	70	50
15 (MI)	Maint.	5;1 M	AI, B,W/H	E (S)	$50K–$74.9K	0	0	87	17 (< 1)	57	9 22	1 20	50	10	10	10	40
16 (I)	Maint.	3;7 M	B/H	E (S)	$50K–$74.9K	0	0	49	20 (8)	74	10 20	5 20	58	45	0	40	10
17 (I)	Maint.	4;5 F	W/NH	E	Over $200K	0	33	187	27 (3)	68	12 33	7 26	83	50	70	80	60
18 (MI)	Maint.	4;4 M	W/NH	E (H)	NR	0	0	114	18 (1)	65	9 21	4 22	65	15	20	40	40
19 (I)	Maint.	4;7 M	B, W/NH	E	$100K–$149.9K	0	0	79	31 (3)	72	8 19	2 21	50	80	30	80	40
20 (I)	Maint.	4;7 M	W/NH	E	$50K–$74.9K	0	6	94	30 (2)	68	10 23	4 23	58	50	0	40	0
21 (I)	Maint.	5;4 F	W/H	E (S)	$25K–$34.9K	0	29	124	13 (< 1)	57	11 31	5 25	83	60	60	50	30
22 (C)	Month 4	3;9 F	W/NH	E	$50K–$74.9K	7	12	117	20 (4)	69	9 19	6 23	83	35	40	30	10
23 (C)	Month 4	3;5 F	W/NH	E	$100K–$149.9K	0	0	47	16 (6)	69	11 25	6 21	65	45	30	20	20
24 (I)	Month 2	5;3 F	W/H	E (S)	$150K–$199.9K	0	0	186	41 (1)	59	10 28	6 26	65	45	10	50	20
25 (I)	Month 1	4;6 F	W, O/H	E (S)	$25K–$34.9K	1	21	106	21 (1)	59	10 22	4 22	65	50	10	20	0
26 (MI)	Maint.	5;7 M	AI/NH	E (T)	$50K–$74.9K	1	0	63	6 (< 1)	40	8 20	1 20	65	0	10	10	0
27 (I)	Month 1	4;4 M	W/NH	E	$15K–$24.9K	0	0	245	24 (2)	68	9 21	6 24	65	45	20	30	0
28 (MI)	Maint.	4;11 F	W/H	E	$50K–$74.9K	0	0	110	3 (< 1)	40	9 23	1 21	41	0	30	20	30
29 (MI)	Month 3	3;4 M	W/H	E (ASL)	$150K–$199.9K	3	0	248	9 (3)	99	13 33	5 19	65	60	90	80	40

Note. I, C, or MI = intervention, control, or monitor + intervention; CA = chronological age; SES = socioeconomic status; AAC = augmentative and alternative communication; I-ASCC = Index of Augmented Speech Comprehensibility in Children, with contextual information; CDI = Communication Development Inventory; TACL = Test for Auditory Comprehension of Language; PPVT = Peabody Picture Vocabulary Test; SS = standard score; VABS = Vineland Adaptive Behavior Scales; AE = age equivalent; Ax = assessment; AAO = agent–action–object; Pos = possessor + entity; Att = attribute + entity; Loc = entity + locative; Maint. = maintenance; M = male; A = Asian; NH = Non-Hispanic; E = English; F = female; W = White; O = other; NR = not reported; S = Spanish; H (ethnicity) = Hispanic; BL = baseline; B = Black; AI = American Indian; H (language) = Hebrew; T = Towa; ASL = American Sign Language.

^aHigh-tech AAC use.

^bCDI = total number of words checked.

^cSymbol Assessment and Comprehension of Semantic Relations are the researchers' own assessments.

Institutional review board (IRB) approval was obtained from the University of New Mexico (UNM) with acceptance and “Rely Upon” IRB approval obtained from by the University of Central Florida (UCF). Consistent with IRB-approved procedures, parental informed consent and child assent procedures were followed. Subsequently, participants completed the assessments listed above, as well as the Peabody Picture Vocabulary Test–Fifth Edition (Dunn, 2018), the Test for Auditory Comprehension of Language–Fourth Edition (Carrow-Woolfolk, 2014), the Vineland Adaptive Behavior Scales–Third Edition (Sparrow et al., 2016), and criterion-referenced assessments designed to examine comprehension skills of the semantic relations targeted during intervention. Procedures for the comprehension of semantic relations are in Table 2. Score calculations on all standardized tests were recalculated by a second person to ensure accuracy, with differences resolved by one of the authors. Participants were not required to have prior aided language experience. No participant experienced any known harm as a result of participating in the study.

Table 2.

Semantic relations comprehension targets and procedures.

Target	No. of video clips	Procedures	Example
Agent–action–object	20	An animal puppet performs an action while the child is asked a yes/no question about the action or the object. In 50% of videos, the correct answer is yes. In the other 50% of videos, the correct answer is no.	Target: Dog is washing the airplane Video clip: A dog puppet is shown on the screen washing a toy airplane with a small rag. Prompt: “Is the dog washing the car?”
Attribute–entity	10	Four small plastic toy animals appear in each video clip with movement and sound, including one target and three foils. After the child watches each video, the examiner asks the child to point to the target.	Target: Red dog Video clip: Small popsicle-mounted plastic red dog, blue dog, red lion, and blue lion toys appear on the screen with movement and sound. Prompt: “Show me the red dog.”
Entity–locative	10	Three small plastic animals—two target animals and one different animal—are placed in various locations around an object. After the child watches each video, the examiner asks the child to point to the target.	Target: Dog is on the plane Video clip: A video panning a toy airplane with one small plastic dog placed inside the plane, another plastic dog placed on the plane, and a plastic lion placed on the plane. Prompt: “Show me, ‘The dog is on the plane.’”
Possessor–entity	10	Two animal puppets appear on the screen, and each animal then picks up two food items sequentially and places them on their respective plates. After the child watches each video, the examiner asks the child to point to the target.	Target: Dog's grapes Video clip: A dog puppet and a lion puppet are shown on the screen while each picks up a banana and then picks up grapes and put them on plates in front of them. Prompt: “Show me the dog's grapes.”

Note. UCF UNM AAC Collaboration. (2025, January 27). Comprehension of semantic relations [Video]. YouTube. https://youtu.be/O7yXEzwJqlg.

Research Design

This study was originally designed as a classic RCT; adjustments to study design were made mid-investigation in response to the effects of COVID-19 shutdowns. The original plan was to enroll 60 participants, including 30 in an intervention group and 30 in a control group, to answer the research questions. However, the COVID-19 pandemic shutdowns interrupted the study and brought in-person service delivery to an abrupt halt for an extended period. In turn, this yielded attrition with enrolled subjects and a pause in enrollment. Participants who had been enrolled leading up to COVID-19 were informed of the government-mandated shutdowns, yielding clinic closures and service delivery halts through a phone call from each site's respective project manager. The iPads with AAC applications remained with the participants during the shutdown for the children's ongoing use, and the families were informed that they were welcome to contact the project managers for any queries or requested support during the clinic closure. Upon research reopening at the project sites, previously enrolled participants were contacted to conduct close-out for their participation in the investigation given challenges with the extended time during which the intervention protocol had to be paused. Referrals were made for these children and their families to access other local AAC services. Recruitment for additional participants to be newly enrolled was reinitiated simultaneously. Ultimately, 29 participants completed the initial baseline measurement session (see Figure 1). These 29 participants enrolled between February 2019 and March 2020 prior to pandemic closures (16 participants, plus an additional participant who did not complete baseline) and between September 2021 and July 2022 (13 participants). The majority of the attrition was due to pandemic shutdowns, with an additional three participants not completing the study for other reasons (one moved, one had medical issues, and one was no longer able to travel to the study site). Data collection was completed in June 2023, at the end of the no-cost extension year of this National Institutes of Health–funded project. Profiles for the 29 participants are in Table 1. Notably, the participant group was racially and ethnically diverse; the investigation included participants self-identifying across all U.S. Census ethnic categories and approximately 40% of participants self-identifying as Hispanic. This participant pool is responsive to calls to address health inequities and to ensure that developed interventions are broadly accessible and effective across populations, including the World Health Organization's (2024) recent guidance on best practices for clinical trials.

Following completion of all assessments, at least one family member of each participant completed a 4-hr AAC technology instructional workshop (details below). After participating in the family member workshop, random group assignment was completed for each participant. Throughout the investigation, group assignment was completed by staff at the opposite study site; that is, the UCF completed participant group assignment for all enrolled participants at the UNM and vice versa. The project managers at each site used an online random number generator to assign participant groups and then communicated group assignment list updates to the project manager at the other site. The only restriction on the randomization lists was that a total of 30 participants were to be assigned to each group. At the outset of the investigation, random group assignment was made to the control or to the intervention group using deidentified participant numbers (e.g., Participant 1 to control, Participant 2 to intervention, etc.). Following the shift in research design in response to the COVID-19 closures, participants were randomly assigned to the intervention group or the monitoring + intervention group (with the latter approach used to maximize participation for each child, given limited time and finances left in the project). Participants assigned to the monitoring + intervention group started the investigation as control group participants and then transitioned to serving as intervention group participants.

Performance on the outcome measure was assessed monthly for both intervention and control group participants. Intervention group participants also received 4 months of twice-weekly intervention sessions (28 sessions total). Control group participants who were enrolled in the study prior to the pandemic did not receive intervention as part of the study; they were encouraged to follow up with the project managers for AAC guidance as needed, which in most cases occurred only as follow-up general questions from parents while the children participated in the monthly measurement session appointments. Maintenance data were collected 3 months after the Month 4 measurement session.

Outcome Measure

The outcome measure was W-MLUSym. Like MLUm, W-MLUSym is designed to be a general measure of grammatical growth, using aided language instead of spoken language (Lee et al., 2022). As detailed in Table 3, to calculate this measure, a W-MLUSym score is assigned to each utterance, and the scores are then averaged for a given session. W-MLUSym scores are calculated by multiplying the number of symbols in an utterance that are relevant to the context by the word order score. Each utterance receives a word order score of 0 (no discernible word order), 0.5 (some word order present), or 1 (no apparent word order errors; Lee et al., 2022). For example, if a child produces COW CAR WASH instead of COW WASH CAR and all symbols are relevant to the context, the utterance is credited with three relevant symbols and earns a word order score of 0.5 (COW is in the correct order, but WASH and CAR are not), so this utterance receives a W-MLUSym score of 1.5 (3 relevant symbols × 0.5 word order score = 1.5 W-MLUSym score). Further examples are included in Table 4.

Table 3.

Weighted mean length of utterance in symbols (W-MLUSym) calculation variables.

Variable	Description
Relevant symbol count	Total count of symbols relevant to the target/context in each utterance
Word order score	Assigned accuracy score rating for each utterance: 0 = No discernible word order apparent 0.5 = Some word order apparent, but not entirely accurate or clear 1.0 = No discernible word order errors

Note. W-MLUSym calculation = number of relevant symbols × word order score.

Table 4.

Weighted mean length of utterance in symbols (W-MLUSym) sample calculations.

Target utterance	Utterance	No. of relevant symbols	Word order score	W-MLUSym score
I EAT CAKE	EAT CAKE	2	1	2.0
COW WASH RED CAR	COW CAR WASH	3	0.5	1.5
COW WASH RED CAR	COW WASH CAR	3	1.0	3.0
LION'S CAR	LION'S TRAIN	2	1.0	2.0
Total				(2 + 1.5 + 3.0 + 2.0)/4 = 2.1

Setting and Experimenters

Sessions were administered by the project managers (including the third author) and by the master's level speech-language pathology students under their supervision. Across the two study sites, a total of 26 different examiners for measurement and play sessions were involved in the investigation, including one project manager at UCF (the third author) and three sequential project managers at UNM. All examiners were required to meet a rigorous procedural standard prior to conducting experimental sessions (see Appendix B). All sessions were conducted in private rooms in university clinics at UCF and UNM. Intervention sessions were scheduled twice weekly. The examiner and child were seated on the floor or at a table, depending on the child's needs and the nature of the investigation tasks in each session.

Targets, Materials, and Instrumentation

The main intervention target for this study was the production of aided rule-based utterances; that is, phrases, clauses, and sentences that adhered to the rules of spoken English. To build diversity of utterances, more specific intervention targets included four semantic relations that could be expanded and combined into longer utterances. These were the same semantic relations included in the comprehension task described above: attribute–entity (BLUE DOG), entity–locative (DOG IN CAR), possessor–entity (DOG-'S CAR), and agent–action–object (DOG DRIVE CAR). Early productions of semantic relations are notoriously difficult to code, as these utterances lack sufficient grammatical information (Lee et al., 2022), so child productions were not coded for specific semantic relations.

The same play materials were used for the same nine play routines across the two study sites. Play routines included vet, vehicles, farm, dessert, birthday party, bedtime, grocery shopping, hide and seek, and lunch. Typical symbolic play materials were selected for each routine, such as various types of automobiles for the vehicle routine. The materials were modified to promote the production of particular structures, particularly adjectives. That is, both big and little versions of many items were included, and many items were painted different colors (e.g., vehicles), placed in different color shirts (for the plush puppets), or covered in glued-on dirt. This promoted productions such as BIG CAR, BLUE DOG, DIRTY AIRPLANE, and so forth. Additionally, images of the faces of each participant and examiner were placed on small and large wooden figurines to facilitate productions of I and YOU. For example, the child could place the figurine of himself in a car and then say, “I AM IN CAR.”

All participants used an iPad¹ with the Proloquo2Go² AAC application to complete the study activities. This included an iPad for home use, which contained the full AAC application with a robust vocabulary, as well as the iPads used during intervention and measurement sessions, which contained activity-specific displays designed for each play routine. For each play routine, four displays were created, with the first display containing a limited number of symbols and each successive display containing additional symbols. Importantly, even the most basic-level display (see “Step 1 display” in Figure 2) contained symbols representing various word classes (i.e., nouns, verbs, adjectives, and prepositions), which is in keeping with using a typical developmental model to guide intervention practices (Binger et al., 2024). As participants progressed, they would “step up” to the next display as they demonstrated the use of the various symbols. Specifically, they moved from the Step 1 to Step 2 display after they used at least one symbol from each word class, from Step 2 to Step 3 after using at least two symbols from each word class, and from Step 3 to Step 4 after using at least three symbols from each word class in two out of three consecutive measurement or intervention sessions.

Figure 2.

Example communication displays.

Notably, as participants stepped up to displays containing additional symbols, the symbols on the original displays remained on the displays in the same position. For example, the dog in Figure 2 remains in the same location on all four displays. Additionally, some symbols appeared on the displays for multiple play routines. For example, the same five animals, five prepositions, and two articles remained the same for every play routine, although particular symbols that appeared on the more limited Step 1, 2, and 3 displays varied according to the needs of each play routine. Some of the verbs and nouns changed according to the routine; for example, for the vehicle routine, displays contained vehicles such as TRAIN and verbs such as CRASH, and for the dessert routine, displays contained dessert items such as COOKIE and verbs such as BAKE.

All sessions were recorded using the Noldus system recording equipment, which includes two permanently mounted cameras in each recording room. The recordings from each of the two cameras are synchronized, allowing for multiple viewpoints during behavioral data analysis, including full room view and zoomed in view of the AAC device displays. All data were coded and analyzed using Noldus's Observer XT software program, which synchronizes the two videos, along with the coding scheme and supports the use of multifaceted codes (e.g., coding symbol relevance as well as word order scores).

Procedure

Caregiver AAC Technology Instructional Workshop

At least one caregiver for all participants attended a 4-hr hands-on AAC technology instructional workshop prior to the baseline measurement session. This represented a “Business as Usual” or “Standard of Care” approach to including a “Control Group” and is in line with clinical procedures when children receive iPads with AAC apps. Participants in the control group only received this Standard of Care caregiver/parent workshop, without AAC-GLI implemented with the children during the control period. The vast majority of these instructional workshops were administered by a project manager, with a few sessions administered by a trained upper-level graduate student. The instructional session included an introduction and general definitions and applications of (a) assistive technology and AAC; (b) AAC and speech-language development; (c) basic operations of an iPad, programming of the Proloquo2Go application; and (d) language facilitation techniques such as aided modeling, wait time, and adding information to child-produced utterances (i.e., expansions and extensions). Practice exercises and video exemplars were used during the workshop.

Measurement Sessions

All outcome measure data were collected during play-based measurement sessions. All participants completed an initial baseline measurement session, and subsequent measurement sessions took place approximately 1 month later. For the intervention group participants, these sessions took place after they completed seven intervention sessions (i.e., after their seventh, 14th, 21st, and 28th intervention sessions). The only exceptions were if a participant missed numerous sessions; for example, one participant was hospitalized for a period of time, missed a measurement session, and resumed measurement in a subsequent month. Measurement sessions were conducted in the same manner for all participants. The sessions began with a brief review of the symbols that would be used during the session, such as the symbols in Figure 2, followed by a 25-min play session. As a general rule, measurement session procedures largely mirrored the intervention sessions described below (and using the techniques in Appendix A), with one significant exception: In the measurement sessions, the examiners did not use techniques designed to elicit, correct, or co-construct specific structures. They could not, for example, provide a correction after the child produced UNDER CAR by saying, I think you meant Dog is in the car DOG IN CAR. Caregivers also completed a brief monthly survey at the time of each measurement session to provide information on the children's perceived communication at home during the preceding month and any other desired caregiver/parent anecdotal comments.

Intervention Sessions

AAC-GLI was implemented throughout the investigation. As described in the introduction, AAC-GLI focuses on three primary components: intervention techniques, technology implementation, and contexts. The technology and contextual components were the same for the intervention and control groups and are described above in the Measurement Sessions section. That is, iPads with an AAC app programmed with activity-specific displays such as those in Figure 2 were used in intervention sessions, and the play routines described above were used as contexts for intervention (vet, vehicles, etc.). The same play routines were used for up to two consecutive sessions and were changed thereafter to ensure exposure to a range of vocabulary. The focus was on participants learning to adhere to the underlying linguistic rules (e.g., agents are followed by actions and then objects) rather than on memorizing specific symbol combinations.

The AAC-GLI approach is rooted in a developmental model of aided language development, with an array of developmentally appropriate syntactic and grammatical structures targeted while simultaneously working to improve expressive semantic skills (Binger et al., 2024). The focus is on teaching productions of the underlying linguistic rules rather than memorizing particular symbol combinations. AAC-GLI intervention techniques are detailed in Appendix A, including links to videos that exemplify each technique. The techniques that examiners used at any given point in time varied according to the needs of each particular interaction. For example, the examiner might begin with an open-ended spoken question (What animals do you want to play with?), and if the child did not answer the question, the examiner might then provide a spoken and aided binary choice (Do you want the big car BIG CAR or the little car LITTLE CAR?). Within every session, the examiner was required to model agent–action–object and entity–locative relations using first-, second-, and third-person subjects at least twice. For example, for agent–action–object, exemplars might include I WASH CAR, I DRIVE CAR, YOU WASH CAR, YOU DRIVE CAR, DOG WASH CAR, and DOG DRIVE CAR. The exception was for participants using Step 1 displays (see Figure 2), which lacked the pronoun YOU. For attribute–entity and possessor–entity targets, the examiner was required to provide at least two models of each relation (DIRTY DOG, BIG HIPPO; DOG-'S CAR, HIPPO-'S TRAIN). Beyond these requirements, examiners typically focused on one or two relations (typically two) during each session using a range of vocabulary; that is, using many repetitions with variety, such as DIRTY DOG, DIRTY HIPPO, BIG DOG, BIG HIPPO, LITTLE DOG, LITTLE HIPPO, and so forth. The targeted relations varied across sessions to ensure exposure to all targets.

Data Collection and Analysis

All sessions were video-recorded using the two-camera system by Noldus as described above. Measurement sessions were analyzed using Noldus's Observer XT software. We used Observer to develop a behavioral coding system in which a variety of codes were used to characterize each utterance. Before assigning codes, utterance boundaries first had to be determined. Participants were taught to select the message bar at the top of the screen once they completed an utterance (the top of the displays in Figure 1 shows the message bar). In general, final utterances that appeared on the message bar were the utterances that were transcribed, as long as all symbols were selected by the participant and not the examiner. This allowed the child to self-correct while constructing messages, and only the final utterance was used for analyses.

Once an utterance was transcribed, it could be coded. Code categories relevant to the current study included the following: (a) communicative intent, (b) listing, (c) imitative versus nonimitative, (d) independent versus co-constructed, (e) number of relevant symbols, and (f) word order score. Description of each of these code categories can be found in the coding scheme in Table 5. Utterances that were included in outcome measure calculations for W-MLUSym had communicative intent, did not consist of simple lists of symbols, and were produced independently.

Table 5.

Augmentative and Alternative Communication Generative Language Intervention coding scheme.

Category	Description
Communicative intent	Communicative intent was coded as present if the child produced an utterance for which any part was judged to be a meaningful intent to communicate (compared with utterances in which the child randomly selected symbols). Utterances with no communicative intent were excluded from subsequent analyses.
Listing	Utterances were coded as listing if the production consisted of a sequence of three or more different symbols with the same part of speech or category (e.g., COW DOG DUCK) and did not include any kind of grammatical information. For example, COW DOG DUCK would be coded as listing, whereas DRIVE COW DOG DUCK would not be coded as listing. Listing utterances were excluded from additional analyses as these utterances can inflate W-MLUSym and MLUSym.
Imitative vs. nonimitative	Utterances were coded as imitative when they included only words or symbols present in the examiner's immediately prior utterance; all other utterances were categorized as nonimitative.
Independent vs. co-constructed	Utterances were coded as independent utterances were produced without substantial prompting from the instructor, and in contrast, co-constructed utterances were produced with assistance. Independent utterances could receive operational supports (such as help with erasing a message or with selecting the message bar), but not linguistic supports (e.g., the examiner selects or points directly to one or more symbols). If the examiner provided spoken prompts, even if they did not tell the child which particular symbol to select, or if the examiner selected or pointed to a particular symbol, the utterance was coded as co-constructed. Only utterances that were independently constructed by the child were included in additional analyses
No. of relevant symbols	The number of relevant symbols for each utterance was required for calculating W-MLUSym and was determined based on contextual information. In general, participants were given the benefit of the doubt. For example, nouns were considered relevant if they were mentioned in a recent conversational turn, were within view of the child, or were being requested or were commented upon by the child. Repeated symbols were only counted once; for example, DOG DOG DRIVE CAR was considered to contain three symbols, not four. Symbols that were clearly selected in error while searching for a particular symbol were not counted as relevant.
Word order score	Word order score was a measure for word order accuracy of each utterance. Each utterance receives a score of 0, 0.5, or 1. 0 = No discernible word order is apparent. 0.5 = Some word order is apparent but is not entirely accurate or clear. 1.0 = Word order has no discernible errors.

Note. W-MLUSym = weighted mean length of utterance in symbols; MLUSym = mean length of utterance in symbols.

Group differences on W-MLUSym were determined by using Hedges's g effect sizes (Hedges, 1981). Generally accepted interpretation of Hedges's g effect sizes (Hedges, 1981) follows that of Cohen's d (Cohen, 1988) and are as follows: 0.2 is a small effect, 0.5 is a moderate effect, and 0.8 is a large effect.

Measurement and Intervention Fidelity

Fidelity measures were completed for both the measurement and intervention sessions to ensure accurate implementation of study procedures. Communication disorders students at both UCF and the UNM completed all fidelity checks. Fidelity checks were completed for at least 20% of the measurement and intervention sessions for each participant. To minimize biases, students at UCF completed fidelity checks for UNM sessions and vice versa. If the students from the same site coded these data, they did not know or have contact with the graduate students or participants who conducted the original intervention sessions. Coders were masked from the order of the participant sessions; all coding was completed by viewing randomly ordered video-recorded sessions. Fidelity indicators for measurement and intervention sessions were the same except for the exclusion of some intervention techniques during measurement sessions (see Appendix A). These indicators included the following: session length of at least 24 min, access to iPad with appropriate display, completion of symbol review, use of a range of intervention techniques, use of grammatically complete spoken models (as opposed to using telegraphic speech), and at least two correct aided productions each by either the examiner or child for the following semantic relations: attribute–entity; possessor–entity; first-, second-, and third-person agent–action–object; and first-, second-, and third-person entity–locatives. Detailed data were collected to assess the intervention techniques used as well as the various aided productions. Additionally, the examiner's aided utterances were required to average between 0.5 and 2.0 symbols longer than the mean of the child's aided productions, and examiners were to select the message bar after each of their aided utterances (which resulted in playback of the entire utterance). Procedural fidelity measures were calculated for at least 20% of the measurement sessions for each child and at least 20% of the intervention sessions for each child. The overall mean measurement session fidelity was 91% (range: 64%–100%) for measurement sessions and 95% for intervention sessions (range: 60%–100%). Fidelity measurement fell below 80% for several sessions when a new project manager began working at the UNM site, but adherence to the procedural protocol rapidly recovered. Overall, the results indicate consistent administration of procedures.

Data Reliability

Communication disorders students at both study sites viewed the same randomly ordered measurement sessions that were used for interrater reliability purposes. As with fidelity, students at UCF viewed UNM sessions whenever possible, and vice versa. Reliability for W-MLUSym was examined. Intraclass correlation coefficients (ICCs)—specifically, ICC2, in which both raters and participants are viewed as random—were used to calculate agreement scores. ICC2 was .81, which is considered good agreement (Koo & Li, 2016). The data used for the results are therefore considered to be reliable.

Results

Pre-Intervention Group Comparisons

The intervention and control groups were compared to assess possible group differences on various language assessments (see Table 6) and a speech assessment (see Table 7) prior to scheduling the first measurement session. Hedges's g was calculated to examine the effect sizes of group differences (Hedges, 1981). Weak, insignificant differences were noted on all variables. Overall, the findings indicate that the groups were well matched in terms of the speech and language measures conducted at the onset of enrollment.

Table 6.

Group equivalence variable means (standard deviations) at baseline by intervention condition.

Variable	Intervention	Control	Hedges's g (CI)a	p b
Chronological age (months)	54.45 (8.689)	50.86 (8.094)	0.415 [−0.250, 1.075]	.108
Female/male ratioc	6:8 (42% F)	7:15 (32% F)
I-ASCC % correct	9.73 (13.484)	10.29 (13.561)	−0.041 [−0.695, 0.615]	.452
CDI total no. words	141.73 (87.960)	143.00 (85.134	−0.014 [−0.669, 0.641]	.483
TACL-4 SS	6.182 (8.234)	6.179 (8.446)	< 0.001 [−0.655, 0.656]	.500
PPVT-5 SS	31.14 (2.396)	31.50 (3.132)	−0.132 [−0.787, 0.525]	.348
VABS-3 Communication SS	62.95 (8.925)	66.93 (10.141)	−0.413 [−1.072, 0.252]	.113
Basic comp: AAO % correct	45.45 (24.246)	41.786 (26.210)	0.143 [−0.514, 0.798]	.335
Basic comp: Pos % correct	43.636 (34.578)	50.00 (32.106)	−0.185 [−0.840, 0.473]	.292
Basic comp: Att % correct	51.364 (26.421)	41.429 (31.343)	0.342 [−0.321, 1.000]	.157
Basic comp: Loc % correct	30.455 (24.197)	31.429 (25.071)	−0.039 [−0.694, 0.617]	.454

Note. Intervention n = 22, control n = 14. CI = confidence interval; F = female; I-ASCC = Index of Augmented Speech Comprehensibility in Children; CDI = MacArthur–Bates Communication Development Inventory; TACL-4 SS = Test of Auditory Comprehension of Language–Fourth Edition standard score; PPVT-5 SS = Peabody Picture Vocabulary Test–Fifth Edition standard score; VABS Communication SS = Vineland Adaptive Behavior Scales–Third Edition Communication subtest standard score; Basic comp AAO, Pos, Att, & Loc = Basic comprehension task for agent–action–object, possessor–entity, attribute–entity, and entity–locative, respectively.

^aHedges's g effect size is interpreted as the difference between means in standard deviation units; the denominator is the sample standard deviation of the mean difference adjusted by a small sample correction factor. Positive g indicates intervention condition has the higher average.

^bObserved probability value (p) is estimated from an independent t test and is based on a directional hypothesis (i.e., one-sided p).

^cChi-square test of association suggests there is not a statistically significant association between treatment condition and gender with a small effect based on the phi coefficient, χ² = 0.452, p = .501, phi = −0.112.

Table 7.

Participant Index of Augmented Speech Comprehensibility in Children (I-ASCC; Dowden, 1997) scores.

Group	Preparticipation Group M (SD)	Postparticipation M (SD)
Intervention group	13.18% (11.57%)	19.55% (19.19%)
Control group	8.14% (9.77%)	16.09% (18.87%)

Note. Scores presented for the “With Contextual Information” I-ASCC condition.

W-MLUSym Findings

The W-MLUSym findings for the intervention versus control groups at each measurement point are in Table 8 and are graphed in Figure 3. Several key findings are apparent in Table 8. Most importantly, moderate-to-large effects are apparent by the Month 3 and Month 4 measurement sessions (see Hedges's g data points that exceed 0.7), clearly indicating a higher W-MLUSym for the intervention group. Second, the smaller enrollment and attrition that occurred due to the pandemic affected power. As seen in Figure 1, all participants did not complete all measurement sessions, primarily due to the COVID-19 closures. This, combined with overall reduced group sizes due to pandemic closures, impacted the findings. Generally speaking, power of .8 is considered acceptable. The study was severely underpowered at Month 1 and Month 2. By Month 3, as differences between the groups grew, power levels increased to .638, with moderate-to-strong effects present (g = 0.731). At Month 4, differences between the intervention and control groups were so large—despite having only six controls at this point—that power was nearly at .8, and large effects in favor of the intervention group are present (g = 0.845). Statistical significance at Months 3 and 4 (p = .026 and .012, respectively) further support the superiority of the intervention group's W-MLUSym performance. Maintenance data should be interpreted with particular caution, as only three control group participants completed the maintenance phase.

Table 8.

Weighted mean length of utterance in symbols effect sizes and power by measurement session.

Time	Intervention group n	Control group n	One-sided pa	Hedges's g (95% CI)b	Power
Baseline	22	14	.027	0.826 [−0.136, −1.504]	.637
Month 1	20	14	.409	−0.079 [0.589, −0.745]	.077
Month 2	17	12	.165	0.364 [0.364, −1.085]	.231
Month 3	15	14	.026	0.731 [0.009, −1.459]	.638
Month 4	13	6	.012	0.845 [0.132, −1.800]	.767
Maintenance	13	3	.032	1.213 [0.074, −2.461]	.969

Note. CI = confidence interval.

^aAssumes intervention group is higher.

^bEqual variances not assumed.

Figure 3.

Intervention and control group performance on weighted mean length of utterance in symbols (W-MLUSym).

Discussion

Effects of AAC-GLI on Expressive Aided Grammatical Growth

The most important finding of this study was the superior growth in W-MLUSym for the intervention participants compared with the control participants. By the third and fourth months, the intervention group produced significantly longer utterances than the control group, even given the enrollment and attrition challenges experienced during the COVID-19 pandemic. This finding (a) demonstrates the efficacy of AAC-GLI for supporting the early grammatical development of young children with Down syndrome and (b) emphasizes that offering only a brief caregiver/parent workshop on AAC operation and implementation to support expressive language does not maximize communication gains for children with Down syndrome.

At baseline, the intervention group had a W-MLUSym of 1.10, with little growth apparent for the first 2 months. Yet, this initial limited growth in W-MLUSym was not indicative of the intervention participants not learning; as demonstrated in other AAC studies including children with Down syndrome (e.g., Romski & Sevcik, 1996), measured gains are not always immediately apparent, but can later become apparent. In the early sessions of this investigation, participants were exposed to many aided models and produced many co-constructed utterances, which were not included in W-MLUSym calculations. Then, by Month 3, the instruction and co-construction practice began to yield substantial gains; furthermore, by Month 4, W-MLUSym increased to 1.37 (up from 1.32 at Month 3). Importantly, improvements were noted at each point from Month 1 to Month 4, and even when the study concluded, W-MLUSym results were still trending upward. The gains are impressive in the context of the slow growth rates in spoken mean length of utterance (MLU) known to occur in children with Down syndrome (Chapman et al., 1998; Marques & Limongi, 2011). Given our limited knowledge of how W-MLUSym relates to MLU, comparing these two measures is somewhat challenging. Even so, we can still draw some initial parallels. Spoken MLU for children ages 3–5 years with Down syndrome is, depending on the source, reported to be approximately 1.2 or 1.3 (Aminian et al., 2023; Harris, 1983). Importantly, one of these studies noted the substantial percentage (41%) of the children's productions that could not be analyzed due to poor intelligibility, compared with 11% of the nonverbal age-matched peers. Berglund et al. (2001) presented data from a large pool of children with Down syndrome indicating that at least 25% of these children from ages 1 to 5 years did not yet combine words, and at age 4 years, only approximately half produced two-word utterances.

Notably, the intervention group had significantly greater W-MLUSym scores than the control group at baseline (g = 0.826, p = .027). The intervention group did not have any apparent inherent quantifiable linguistic advantages over the control group (see Table 5), so several other factors might have contributed to this finding. First, the study was underpowered at baseline (see Table 6), meaning that individual differences could have played a role. Second, four control group participants earned a W-MLUSym of “0” at baseline, due to producing no aided utterances or producing utterances with no discernible word order. These “zero” W-MLUSym scores affected the baseline mean for the control group. Notably, the “monitoring + intervention” participants who enrolled after the pandemic closures completed several measurement sessions (during their monitoring period) before completing their intervention baseline session. This was observed to affect their performance; W-MLUSym in their original baseline sessions (i.e., when they were in the control condition) was lower than their second baseline session immediately prior to starting intervention (M = 0.76 and 1.11, respectively). In other words, AAC practice from the measurement sessions and possibly at home application of what was learned during the instructional workshop, in addition to possible maturational factors, likely resulted in higher intervention baseline performance. That is, some of the adjustments made to the study resulting from pandemic realities seemed to have inflated the intervention group's baseline data. Notably, the vast majority of the participants' utterances were nonimitative: 84% for controls and 88% for the intervention group.

Crucially, use of aided AAC bypasses intelligibility issues, which may be partly to blame for the expressive grammar weaknesses that are often noted in Down syndrome (Chapman et al., 1998; Draghi & Zampini, 2019). Given this, the current findings suggest great promise for aided AAC to support the expressive language of children with Down syndrome, particularly given the speed with which the participants began combining words in the current study—and doing so with relevant vocabulary using the rules of spoken word order. In addition to these gains, all 13 of the intervention group participants who completed Month 4 measurements also participated in a maintenance session 3 months later and had maintained their W-MLUSym gains, despite having no additional intervention. Critically, however, they did not demonstrate improvements, indicating the need for ongoing intervention to promote continued gains.

For the control group, the baseline mean was well below 1.0, which, as discussed in the Results section, was largely due to four control group participants having a W-MLUSym of zero at baseline. This phenomenon did not occur with the intervention group at baseline, likely due in part to the fact that the seven “monitoring + intervention” participants had already gained aided language exposure by the time they joined the intervention group. At Month 1, the control and intervention groups had similar W-MLUSym means (i.e., 1.03 and 1.06, respectively). After this, however, the two groups diverged for the remainder of the study, with the control group plateauing. Unfortunately, so few control participants completed the maintenance phase (only three, due to postpandemic changes) that group comparisons must be made with extreme caution. Nevertheless, the precipitous drop in W-MLUSym for these three controls at the maintenance point is notable.

Clinical Implications

The main finding that AAC-GLI can be used to teach young children with Down syndrome how to progressively build rule-based phrases and clauses is highly encouraging and opens new possibilities for the use of AAC with this population. This finding supports the effectiveness of using AAC to go beyond simply supporting pragmatic and semantic growth (Barton-Hulsey et al., 2021; Romski et al., 2023). To our knowledge, this is the largest study to date, as well as the only RCT that explores aided language grammatical growth in children with Down syndrome. As reported in depth elsewhere recently (Binger et al., 2024), implementation of a developmental model of language development forms an evidence-based foundation to ensure that all relevant aspects of language are considered when planning aided intervention; developing syntactic and grammatical skills is no less important for children with Down syndrome than it is for other children. As the results of this investigation indicate, aided AAC can be used to support essential syntactic and grammatical skill development.

For a host of reasons, some families and educators may be reticent to seek AAC solutions for young children with Down syndrome—such as beliefs that the child will soon become more intelligible, or that they will primarily use speech to communication—thus negating the need for AAC. None of these reasons, however, support the failure to provide AAC solutions and services.

First, although intelligibility is likely to improve over time, intelligibility is problematic throughout the life span for many people with Down syndrome (Berglund et al., 2001). When individuals do primarily rely on speech, AAC can still be effective to solve communication breakdowns (Beukelman & Light, 2020), and early experiences with AAC should facilitate such use. Furthermore, substantial data have demonstrated that use of aided AAC does not hinder the use of speech (Romski et al., 2023). When children experience communication success with speech, they continue to use spoken communication. A full analysis of the caregiver/parent survey data obtained during monthly measurements sessions in the present study is beyond the scope of the present article; however, anecdotal written feedback provided within these monthly surveys reflected simultaneous speech success when parents expressed excitement about their children using the grammatical structures they learned to use in their AAC intervention sessions in their everyday speech at home and school as well. Parents further reported a sense of gratitude for their children having the benefit of participation in the study and relief in knowing their children had a back-up mode of communication in instances where their natural speech was not understood in school and community contexts. We interpret this type of open-ended written feedback as high praise, particularly from this group of parents, given that very few of the participants had used AAC prior to enrolling in the study.

The question, then, should not be “Why should we use AAC?” but instead should be “Why aren't we already using AAC?” Recent findings from Romski et al. (2023) demonstrating the benefits of aided AAC in supporting early vocabulary growth combined with the results of the current study provide a strong case for providing aided AAC services—which include research-based intervention approaches such as AAC-GLI—as a standard part of early intervention, preschool, and school intervention programs for children with Down syndrome. That is, we begin to have research-based guidance for educators, families, and clinicians on how to support early expressive semantic and grammatical development, and these solutions need to be integrated into school- and clinic-based services. Additionally, as we know from the literature on developmental language disorders, spoken language skills underlie success in literacy (Paul et al., 2018); therefore, investing in these foundational skills is likely to have far-reaching positive consequences. Such investment undoubtedly must involve advocating for skilled AAC service delivery and supports as Individual Family Service Plans and Individual Education Plans are developed for children with Down syndrome. Without ready access to relevant technologies and language intervention, valuable time can be lost in the language development process, and communication and literacy outcomes may be limited. Parent feedback in the present investigation underlined the sense of relief they felt and the widespread communication and educational benefits they observed in their children during their participation in the investigation.

Limitations and Directions for Future Research

Complications resulting from conducting this RCT in the midst of the COVID-19 pandemic presented obvious limitations discussed in detail in the Results section (e.g., varying numbers of participants at different points in the investigation); in turn, these challenges yielded less-than-ideal power than originally planned for this group design. In addition, the intervention group presented with greater W-MLUSym scores than the control group at baseline, a factor that may not have occurred if the study had been fully powered. Despite these limitations, however, the effects of AAC-GLI were strong enough that the superiority of the intervention group gains in W-MLUSym are still convincing. Having fewer participants did mean, however, that it was not possible to analyze potential moderators of the intervention. For example, we originally hypothesized that stronger receptive language skills would lead to larger W-MLUSym gains, but this will need to be investigated in a future study with increased power given the challenges ultimately encountered in this study. Relatedly, how well the findings generalize to the general population of children ages 3–5 years with Down syndrome is unknown. Better understanding of which children will benefit from an intervention and under what circumstances is of great future interest.

The findings open a host of additional avenues for future research. One pressing question, for example, is how much grammatical progress young children with Down syndrome can make—and how quickly—when using aided AAC for a period of longer than 4 months? Relatedly, further questions remain about lexical growth and diversity. In addition to grammatical benefits, aided AAC also can provide distinct semantic advantages by allowing children to easily “say” phonetically challenging words using voice output. We have just begun to investigate these questions in a new research project in which we will follow the aided language growth of children with Down syndrome (as well as those with other disabilities) for a 2-year period. As noted in Figure 3, the participants in the current study were still making progress at the end of their 4 months of intervention. We would expect this trend to continue for some time, but more work is needed to better understand ongoing aided grammatical growth. A related question stemming from the previously described family feedback is how such growth relates to simultaneous growth in spoken language as well as shifts in levels of intelligibility (Barton-Hulsey et al., 2021). Children with Down syndrome are multimodal communicators, and little is currently known about how different communication modes relate to and support one another.

Ongoing measurement research is needed as well, as we currently know so little about aided language measures. In our current work (Lee et al., 2022), we are investigating a host of promising measures, some of which were adapted from existing spoken language measures (such as W-MLUSym), and some of which are unique to aided AAC (e.g., percentage of relevant symbols). A related issue is the need to better understand the nature of W-MLUSym and how it relates—and does not relate—to spoken MLU. We already know, for example, that these measures do not mean the same thing; W-MLUSym, for example, can be less than 1.0, which is not true of MLU. Systematic inquiry, then, into the validity, reliability, developmental sensitivity, and clinical applicability of these new measures is required, so we can better monitor the language gains of children who use AAC, including children with Down syndrome.

Finally, we are acutely aware that these pressing future research directions should be considered and approached in the context of an implementation framework. Implementation science approaches hold great promise for lines of AAC research such as this one (Olswang & Prelock, 2015). Parents of participants readily discussed the benefits experienced by students when they were provided with AAC intervention and the growth they witnessed in their children's overall communication and educational success. Through deliberate stakeholder group partnerships—including self-advocates, family members, clinicians, researchers, administrators, and policy makers—we are confident we will be able to adapt AAC-GLI for practical use in a variety of settings to better ensure all children with Down syndrome have ready access to evidence-based services to support their communication, social, and educational success.

Conclusions

In conclusion, the findings of the present investigation indicate that only providing children with Down syndrome and their families access to relevant AAC technologies and a general overview of evidence-based AAC applications is not sufficient to yield positive changes in the children's aided expressive language skills. However, encouragingly, the findings do indicate that language-based AAC interventions present great promise to teach young children with Down syndrome to improve their aided expressive grammar skills. That is, providing young children with Down syndrome access to research-based AAC language interventions, such as AAC-GLI, can be a critical step to support ongoing expressive language development and use. With access to both relevant AAC technologies and related language intervention, children with Down syndrome can be well supported to achieve functional communication, literacy, social, and educational success.

Data Availability Statement

Links to videos depicting various intervention techniques are available in Appendix A. A link to the videos used to collect data on the comprehension of semantic relations is at the bottom of Table 2. Additional data are available upon reasonable request from the first author.

Appendix A

AAC-GLI Intervention Techniques

Technique	Rationale	Description or example(s)	Used in measurement sessions?	Used in intervention sessions?
Aided modeling	Direct modeling of morphosyntactic structures aids learning of targets, typically prefaced by a spoken model.	Clinician: “Look! Big Dog is in the car” BIG DOG IN CAR. Always accompany aided utterances with grammatically correct spoken language.	X	X
Binary choice	Use when the child struggles to make choices based on open-ended questions. Use also to highlight word order issues. Can include spoken and aided models.	Clinician: “Should big Dog BIG DOG or dirty Duck DIRTY DUCK help us with our washing?”	X (spoken binary choices only; not word order issues)	X
	Also used to highlight word order issues	Child: DOG BIG Clinician: “Did you mean that you want Dog big or big Dog BIG DOG?”
Contingent responses	Use recasts, expansions, and extensions to respond to the child's communication attempts. Often include spoken and aided models.	Participant: DOG WASH Clinician: “Yes, I see Dog is washing.” DOG IS WASH -ING.	X	X
		Participant: DOG Clinician: “That's big Dog.” BIG DOG
		Participant: DOG CAR Clinician: “Yes, that's Dog's car.” DOG 'S CAR
Describe the action	Provide descriptions of the actions to provide the child with possible vocabulary, prompt aided utterances, and keep the routine going. May include spoken and aided models.	Clinician: “I see Dog washing the car.” DOG WASH	X	X
Direct spoken and/or aided prompt	Use spoken language to tell the participant what to say. Technique is used sparingly.	Clinician: “Tell me, Dog is big.” Clinician: “Tell me, Dog is big. DOG IS BIG. Your turn.”		X
Feign ignorance	Pretend to not understand what the child is trying to say and directs the participant's attention back to the display.	Dog puppet says in Dog's gruff voice: “I know you want me to do something, but I'm not sure what. Will you tell me with the iPad?”	X	X
Open-ended spoken and/or aided questions	Encourage the child to make choices and describe the action of the play routine via the activity-based display.	Clinician: “What would you like to wash today?” or “Hmm, what should we do next?”	X	X
Point to the symbols	Point so symbols on the device to help the child create longer, more complex utterances that currently are too challenging to create independently.	Clinician: “Let's do it together. Dog is big. Dog (point to DOG) is (point to IS) big (point to BIG).”		X
Point to the toys	Draw the child's attention to things they can comment on; this may or may not be accompanied by spoken comments.	Clinician: [Points to the car to indicate that the child might want to select that next].	X	X
Repair utterances	Allows opportunities to fix mistakes, often using spoken and/or aided binary choice	Clinician: “Oh listen! Did you want to say, Dog in car or Dog car in? Oh, Dog in car DOG IN CAR, let's fix it.” Assist the child with deleting the incorrect words as needed.		X
Respond to all communicative attempts	Emphasize aided communication but does not ignore the child's other communicative attempts. Respond and draw his/her attention back to the device.	Clinician: “I'm not sure who you're talking about. Tell me here.” [Point to the display]	X	X
Select the message bar	Encourage the child to select the message bar at the top of the screen once their utterance is complete so they can hear their utterance and self-monitor accurate/inaccurate selections.	Clinician: Say, “Tell me the whole thing” while pointing to the message bar.	X	X
Wait time	Provide the child the opportunity to initiate as well as respond. It also provides crucial processing time. For some children, you can look expectantly at the child, so they know you are excited to hear what they have to say.	Clinician: Hold up the big car and the dirty car, look at the child, and wait for at least 5 seconds.	X	X

Note. From: “Supporting Communication and Language Development in Preschool Children Using AAC,” by N. Harrington, C. Buchanan, J. Kent-Walsh, & C. Binger, in M. M. Smith (Ed.), Clinical Cases in AAC. Copyright © 2023 by Taylor & Francis. Adapted by permission of Taylor & Francis Group.

Appendix B

Personnel Training Checklist

Name:                        Role(s):

	PIs and Project Managers	Examiner/Clinician	Camera Operator	Masked Rater	Date of Completion/	Initials of Core Personnel
Human Research Social / Behavioral Research Investigators and Key Personnel (CITI)	*	*	*	*
Responsible Conduct of Research (RCR through CITI)	*	*	*	*
Good Clinical Practice (GCP): Social and Behavioral Research Best Practices for Clinical Research (Learning Central/CITI)	*	*	*	*
COI (UCF-CITI /UNM-NIH)	*	*	*	*
COI through UNM (UNM only)	*	*	*	*
Review and Discuss: Research design flowchart	*	*
Read and Discuss: Beukelman & Mirenda, Ch. 11 (Supporting Language Development) or Beukelman & Light, Ch. 11 (Intervention to Build Communicative Competence, pp. 375–390.)	*	*
Read and Discuss: “Early Sentence Productions of 3- and 4-Year-Old Children who use AAC” (Binger, Kent-Walsh, King, & Mansfield, 2017)	*	*
Read and Discuss: “Dynamic Assessment for 3- and 4-Year-Old Children Who Use Augmentative and Alternative Communication” (Binger, Kent-Walsh, King, Webb, & Buenviaje, 2017)	*	*
Read and Discuss: “Tracking Early Sentence-Building Progress in Graphic Symbol Communication” (Binger et al., 2020)	*	*
Read and Discuss: “Aided Language measures: Establishing Observer Agreement for Communicators in Early Language Phases” (Lee et al., 2022)	*	*
Video observation PowerPoint (DA, Measurement, and Intervention): View all video clips and answer the associated questions. Must score an 80% or better on the questions.		*
Reading: R01 Standard Operating Procedures for all tasks associated with role(s)	*	*	*	*
Reading: Informed consent form for the project	*
Reading: Leiter manual and protocol	*
Reading: Vineland manual and protocol	*
Reading: MacArthur-Bates CDI Words and Sentences		*
Reading: TACL assessment manual and protocol		*
Role play: Administration of TACL; successfully administer at least 19/20 consecutive items		*
Role play: Administration of Basic Semantic-Syntactic Comprehension task; successfully administer 19/20 consecutive items		*
Role play: Administration of IASCC; successfully administer 19/20 consecutive items		*
Role play: Dynamic Assessment task; successfully administer 9/10 probes		*
Role play: Activity Scene measurement; at least 90% accuracy modeling aided productions of 4 basic structures during this task		*
Role play: Play intervention; at least 90% accuracy on modeling aided utterances of 4 basic structures during play intervention		*
Training: Complete live training on operating cameras			*
Training: Complete live training on coding data for intervention sessions		*	*
Training: Complete training on the coding measurement sessions for Play Mx and Activity Scene sessions				*
Training: Complete training for play intervention data coding				*
Training: Complete Fidelity Training for Ax., Measurement and Intervention Sessions				*

Publisher Note: This article is part of the Forum: Changemakers Igniting Innovation.

Funding Statement

Preliminary data for this article were presented at the 2021, 2022, and 2023 annual conventions of the American Speech-Language-Hearing Association. This research was supported by National Institutes of Health Grant R01DC016321, which was awarded to Principal Investigator Binger and Multiple Principal Investigator Kent-Walsh.