Using AAC to Unlock Communicative Potential in Late-Talking Toddlers

Abstract

Purpose:

The purpose of this study was to report on modifications we made to a standardized input-based word learning treatment for two late-talking toddlers. The modifications were the addition of an augmentative alternative communication (AAC) device and the requirement that the children use this device, or speech, to communicate.

Method:

We used a single-subject design to track late-talking toddlers’ progress through an input-based word learning treatment, which was part of a larger study. Because the input-based treatment protocol was not effective for each toddler based on absent or clinically insignificant treatment effect sizes, we modified the protocol. The modifications were meant to address each child’s potential over-reliance on nonverbal communication and the potential impact of speech sound delay. We then measured their linguistic output.

Results:

Both toddlers showed no evidence of learning during the input-based treatment. Each child’s linguistic output increased by over 600% once we made the protocol modification and introduced the AAC device. They used both AAC and vocal speech to communicate. Both toddlers produced novel words, and one began to produce multiple word combinations.

Discussion:

While input-based therapy has an evidence base and has been successful for some toddlers, it may require modifications for children who have not learned the pragmatic convention of using spoken language, and for children with difficulty with speech sound production.

This paper demonstrates the effect of modifying a standardized input-based language intervention by using an Alternative/Augmentative Communication (AAC) device with two late-talking toddlers. We will describe the standardized treatment, the individual characteristics of each toddler, explain the rationale for the modification, and describe the results. By sharing this information, we hope to encourage clinicians to explore the use of AAC with clients who share characteristics of the toddlers in this study, which include: late language emergence, a receptive vocabulary superior to expressive vocabulary, speech sound production delays, and an over-reliance on nonverbal communication. Although there is a literature that describes the effective use of AAC with toddlers (e.g., Romski, Sevcik, Barton-Hulsey, & Whitmore, 2015; Solomon-Rice & Soto, 2014), often the participants in those studies have more serious developmental delays, and are unlikely to be classified as late-talkers due to other primary diagnoses such as intellectual disability, autism, or apraxia.

Late Talkers

Up to 20% of toddlers are known as ‘late talkers’ (ASHA, “Late Language Emergence,” n.d.; Rescorla, 1989; Rice, Taylor & Zubrick, 2008; Zubrick, Taylor, Rice, & Slegers, 2007). Although specific criteria for identifying late talkers vary (Paul, 1991; Rescorla, 1989; Thal & Bates, 1988), generally, late talkers are children who have reduced expressive vocabulary repertoires and have difficulty producing multi-word combinations in comparison to their age-matched peers. This phenomenon has often been referred to in the literature as late language emergence (LLE), which describes a delay in language acquisition and onset with no other diagnosed developmental delays or disabilities (Zubrick et al., 2007). Typically, a diagnosis of LLE is made when a toddler’s language development is below age-level expectations. For example, well-established language milestones dictate that a child should have a 50-word expressive vocabulary and emerging multi-word combinations by 2 years of age (e.g., ASHA, “One to Two Years,” n.d.; CDC, “Learn The Signs,” n.d.). Children with LLE tend not to meet this milestone.

LLE is worth attending to in early childhood (Law, Garrett, & Nye, 2004; Weismer, Murray-Branch, & Miller, 2004). Although the majority (approximately 50 to 70%) of late-talkers with age-appropriate nonverbal and receptive language skills catch up to their peers by three to four years of age (Dale, Price, Bishop, & Plomin, 2003; Rescorla, Mirak, & Singh, 2000), lingering effects are detectable later in development. For example, 17-year-old children previously identified as late-talkers scored within normal limits, but significantly below the scores of typically-developing peers, on standardized measures of expressive and receptive language abilities (Rescorla, 2009). More importantly, some late-talkers do not catch up to typically-developing peers. They continue to demonstrate language abilities below age-level expectations with normal nonverbal cognitive skills. These children go on to be diagnosed with developmental language disorder (Dale et al., 2003), previously referred to as specific language impairment (Bishop, Snowling, Thompson, & Greenhalgh, 2016; 2017).

Characteristics of Late Talkers

Receptive Language and Speech Sounds.

Although LLE is often defined by poor expressive vocabulary skills and limited language production, research has shown that late-talkers may present with delays specific to expressive language or with mixed expressive and receptive language delays (e.g., Desmarais, Sylvestre, Meyer, Bairati, & Rouleau, 2008). Once they begin to produce words, late-talkers may also demonstrate speech sound delays, with reduced consonant and vowel inventories and less complex syllable structures (Mirak & Rescorla, 1998; Paul & Jennings, 1992). MacRoy-Higgins and Schwartz (2013) also note that toddlers with LLE may have less complex babble than their peers. There is not much known about the speech skills of late talkers, partly because “assessing toddlers’ speech is challenging” (Hodges, Baker, Munro, & McGregor, 2017, p. 587). However, Hodges et al. (2017) undertook this challenge and found that toddlers had different response types during speech assessments including no response, use of protowords, different verbal responses, correct use of phonemes, common phonological errors, and uncommon phonological errors. In this study, toddlers with LLE demonstrated use of all six of these response types, typically at different rates than toddlers with typical language emergence. The contrasts were not favorable for the toddlers with LLE with more non-responses, use of protowords, uncommon errors and fewer correct phonemes. There also appear to be longer-term negative speech outcomes for children who have LLE. Neam, Baker, Hodges, and Munro (2019) found that 4- and 5-year olds with a history of LLE had lower speech accuracy than peers with normal language emergence. In the other direction, toddlers with LLE who had the most difficulty with speech sound production were more likely to have longer-lasting language impairments than toddlers with LLE who had better speech sound production (MacRoy-Higgins & Schwartz, 2013). Speech difficulties may be associated with LLE, at least for a portion of toddlers with LLE.

Prelinguistic Communication.

Another characteristic of late-talkers is that their skill development and pattern of usage of prelinguistic communication differs from typically-developing peers. Before typically-developing toddlers say their first word at approximately 12 months of age, they experience a prelinguistic phase of communication development that is marked by acquisition and use of vocal and nonverbal behaviors, including non-linguistic vocalizations, eye contact, gestures, facial expressions, and turn-taking to engage in social interactions and protoconversations. There are generally three patterns to describe the process of gesture and language emergence: (1) words predominate, (2) words and gestures are used equally, and (3) gestures predominate (Acredolo & Goodwyn, 1990). Thal and Tobias (1992) have shown that late talkers follow the third pattern but that they use significantly more communicative non-linguistic gestures (e.g., pointing) and for a greater variety of communicative purposes than their peers. Late-talkers may rely more heavily upon forms of prelinguistic communication to convey messages with the people in their environments than peers who have the benefit of greater language skills (Thal & Tobias, 1992; Thal, Tobias, & Morrison, 1991). They proposed that late-talkers’ reliance on gestural communication is an attempt to bypass obstacles to oral language acquisition.

Treatment Options.

LLE is a condition whose negative effects begin in early childhood and can be long-lasting. It is well-established that LLE can result in long-term language, literacy, academic, and social difficulties (e.g., Dohmen, Bishop, Chiat & Roy, 2016; Rescorla, 2005; 2009), which can reduce vocational opportunities in adulthood (e.g., Johnson, Beitchman, & Brownlie, 2010). This has necessitated the early identification of LLE and the development of effective interventions to remediate deficits and potentially prevent long-term negative outcomes.

When a child is identified as having LLE, there are two clinically-justified options: watch and wait or provide intervention (Olswang, Rodriguez, & Timler, 1998). Watch and wait is just what it sounds like: watch the child and wait to see if the LLE resolves on its own. Singleton (2018) has cautioned against using this practice citing the need for earlier intervention for children at risk of long-term developmental disorders, and the outdated view that most children with LLE catch up. There is no single, gold-standard intervention approach for children with LLE, but there is evidence supporting both direct (e.g., VAULT, Alt et al., 2020) and indirect treatments (e.g., DeVeney, Hagaman, & Bjornsen, 2017) for this population.

In terms of intervention, AAC is rarely an option considered for children with LLE. First, it is only within the last decade or two that people have deemed AAC an appropriate intervention for toddlers (Cress & Marvin, 2003). As recently as 2014, Solomon-Rice and Soto noted that there was “…a paucity of AAC intervention research…for children under the age of 3.” (p. 204). Although more studies are showing that AAC is a viable choice for toddlers (e.g., Romski et al., 2010, 2015), the studies that exist generally include children who would be perceived as having more severe developmental disabilities than the children labelled as late talkers. For example, the children in Solomon-Rice and Soto’s 2014 study which found that AAC was a viable option for toddlers included a child with a history of extreme prematurity, a child with developmental delays across “all domains” (p. 205) and a child with gross motor ataxia. In Romski et al.’s (2010) randomized comparison of AAC and non-AAC intervention for toddlers, inclusion criteria specified “….a primary etiology other than delayed speech and language impairment…” (p.353). Branson and Demchak’s (2009) review of the literature on AAC use with infants and toddlers noted that the majority of participants in these studies had developmental delays, either specified (e.g., autism, Down syndrome) or unspecified. LLE was not included in the diagnoses listed. Simply put, most clinicians do not think of AAC as a choice for children diagnosed with LLE, neither as a full communication system, nor as a treatment approach, despite ASHA’s recommendation that it “…may be considered as a temporary means of communication for late talkers” (ASHA, “Late Language Emergence”). This may be due to the lack of representation of this approach for children with LLE in the literature, which is mirrored by ASHA’s Practice Portal’s description of AAC as a tool for those with “…significant and complex communication disorders” (ASHA, “Augmentative and Alternative Communication”) and a list of disabilities that does not include LLE (although the site is clear that the list is not exhaustive). It may also be due to a hesitance to incorporate technology into treatment for toddlers based on negative findings associated with additional screen time for toddlers (e.g., Lin, Cherng, Chen, Chen, & Yang, 2015) and professional recommendations to limit screen time (e.g., AAP Council on Communications and Media, 2016).

Given that most treatments for children with LLE do not involve AAC, a typical treatment option for children with LLE is an indirect service delivery model in which parents are trained to modify the ways in which they speak to their children, as in the Hanen Program’s It Takes Two to Talk (e.g., Weitzman, Girolametto, & Drake, 2017). Our team has been developing a direct approach, in which treatment is delivered directly to the child by a clinician. This input-based treatment is based on principles of implicit, statistical learning (e.g., Alt et al., 2012) and rests on the philosophy that rich linguistic input provided to a language-delayed toddler results in improved language output.

In input-based treatment, the focus is on the clinician providing high density language input to the child. Input-based treatment, in general, has the benefit of lowered behavioral expectations for the child because it does not require communicative or verbal output from the child and instead, only requires the child’s attention. Thus, this intervention method is thought to be appropriate for toddlers. In this paper we present a case study of two toddlers, their progress in our standardized, input-based treatment, and their progress following the modifications we made, including the use of AAC. Each stage of treatment and its results will be presented separately.

Methods: VAULT

Participants

All components of the study below were approved by an Institutional Review Board at the University of Arizona. We included two toddlers in our study who will be referred to as LM and CP. Both were monolingual English-speaking males with no previous history of speech language treatment and no known diagnoses. We initially assessed LM at age 2;8 and CP at age 2;0. An overview of all the assessments and treatments can be found in Table 1.

Table 1.

Schedule of intervention for each toddler

Toddler		Pre-Treatment MCDI	Delay: No Treatment	Post-Delay MCDI	16 Sessions VAULT	Immediate Post-Treatment MCDI	No-Treatment	Follow-Up MCDI	Trial therapy
LM	Age	30 months		32 month		34 months		36 months	36 months
	Length		10 weeks; 1 day		7 weeks; 1 day		5 weeks; 1 day		4 sessions: 1 week; 2 days
CP	Age	24 months				26 months		27 months	28 months
	Length			n/a	7 weeks; 1 day		4 weeks; 4 days		3 sessions: 2 weeks

Note: MCDI = MacArthur-Bates CDI Words and Sentences; VAULT = Vocabulary Acquisition and Usage for Late Talkers

LM started in treatment as part of Phase 1 of a larger research study that used the Vocabulary Acquisition Usage for Late Talkers (VAULT) protocol (described in detail in Alt et al., 2020). CP was in the second phase of VAULT. Phase 1 contrasted the number of target words targeted per session and Phase 2 contrasted the length of the utterance provided as a dose, while keeping the other intervention parameters the same. The methodology for both phases of VAULT is based on principles of implicit learning (Alt et al., 2012) and highlights implicit learning achieved through high density input and both linguistic and contextual variability. The VAULT component of the treatment followed single-subject design, with a baseline established prior to treatment, and daily tracking of progress for both target and control words.

For both boys, initial assessment included a test of cognition to rule out intellectual disability, a hearing and vision screening, and at least one measure of vocabulary (i.e., parent checklist). According to parental interviews, each child’s birth, health, and developmental history were unremarkable, with the exception of a noted delay in meeting expressive language milestones. To assess cognition, we administered the Bayley Scales of Infant and Toddler Development - Third Edition (Bayley, 2006). LM’s standard score was 95, indicating normal intelligence for his age and CP’s standard score was 120, indicating above-average intelligence for his age. We attempted a vision and hearing screening, but both children declined to participate. LM’s mother reported that he passed the vision screening at the pediatrician’s office at his 2-year visit. She also reported that he passed his newborn hearing screening. She had no concerns about his hearing. CP’s mother reported that he passed a hearing screening at the pediatrician’s office at 1.5 years old. The team recommended that both parents follow up with the pediatricians regarding a hearing screening, as is recommended for all late talkers, but also had no immediate concerns about either child’s hearing, based on their responsiveness to verbal communication and reported large receptive vocabulary.

In terms of vocabulary, we administered the Expressive Vocabulary Test - Second Edition (Williams, 2007), which requires children to produce a word when shown a picture of an object. LM did not produce any words on this measure. We did not administer this to CP, as he was not yet 30 months of age, the minimum for this standardized measure. We also had both children’s parents fill out the MacArthur Bates Communicative Developmental Inventories -Second Edition: Words and Sentences (MCDI; Fenson et al., 2007). For the MCDI, parents fill out a checklist of words their child says. LM was in the delay condition of our larger study, meaning we monitored his progress for two months before beginning treatment. We obtained MCDI scores at two points prior to beginning treatment. At 30 months, he produced 15 words from the MCDI checklist. Eight weeks later, at 32 months, he produced 22 words. Both of these scores placed him below the 5th percentile for his age. In contrast, his mother reported that he understood over 446 of the 680 words on the MCDI inventory, demonstrating a clear mismatch between receptive and expressive vocabulary. CP produced 10 words at 24 months, which is also below the 5^th percentile for his age. He too, had a much larger receptive language vocabulary, with his parent reporting that he understood 584 of the 680 words on the MCDI inventory.

VAULT Procedures: Stimuli

After establishing that each child was an appropriate candidate on the basis of their expressive vocabulary, we selected 10 pairs of vocabulary terms to be used during treatment. We chose word pairs using several criteria. The first criteria was that the words were on the MCDI. Next, the child needed to understand the word (according to the parent’s report of the words he understood on the MCDI), but not produce it. From this subset of words, we prioritized those words that parents had indicated were of particular relevance to their family. Each family had selected their top 50 words that they would like their child to say. Then, we matched words by category on the MCDI (e.g., clothing, descriptive words), word length in syllables, and item trajectory using Wordbank (Frank, Braginsky, Yurovsky, & Marchman, 2016). Wordbank is an online tool that plots growth curves for words on the MCDI. We did not explicitly select words based on their phonological make-up, given that these are all expected to be early-occurring words, and as such, the words included a wide range of phonemes (See Table 2). Once word pairs were selected, we randomly chose one word in the pair to target in treatment. The other word in the pair served as a control word to ensure that gains in treatment were not due to maturation. An example of a matched pair of words can be found in Figure 1.

Table 2.

Target and control words selected for each toddler

		LM		CP
Pair #	Target	Control	Target	Control
1	Eat	Drink	Truck	Spoon
2	Cat	Dog	Arm	Leg
3	Look	Help	Ball	Book
4	Play	Jump	Hungry	Heavy
5	Car	Truck	Shirt	Pants
6	Star	Slide	Read	Play
7	Rock	Sun	Apple	Cookie
8	Bye	Hi	Bird	Dog
9	Sit	Walk	Block	Toy
10	Cow	Horse	Kitchen	Bathroom

Figure 1:

An example of a matched pair of words and their item trajectories from Wordbank (Frank et al., 2017).

Baseline and Treatment

Once the 20 words (10 pairs) were chosen for each child (see Table 2), we created a PowerPoint slide show on a tablet to probe the child to ensure he was not using these words. During each probe session, an image was displayed to the child, who was then asked to produce the word (e.g., using the prompt, “What is this?”). A baseline was established in which each boy did not produce any of the 20 words over three separate days measured within one to two weeks. Then, we used previously established random orders to determine the order in which the words would be taught.

Once a baseline was established, treatment began. We used a direct, input-based treatment modality with research clinicians as the treatment agents. Research clinicians for LM included two graduate students in speech-language pathology who were directly supervised by certified speech-language pathologists. The dose was an input of a target word delivered in a grammatically correct utterance in a variety of contexts (e.g., playing with different toys, looking at a book). The doses were delivered in varied linguistic contexts. In other words, the input could not be limited to, “That’s a car.” repeated 45 (or 67) times. Input was presented naturalistically, with no special emphasis on the target word. Ideally, the input coincided with the child’s interests and focus of attention. That is, if the child was looking at a horse, the clinician would not be talking about the car.

LM received 45 doses for six words per session, for a total of 270 doses per session. Sessions were 30 minutes long, so the dose density was roughly 9 doses per minute. The dose context was within play. A variety of activities were used for each target word, both within and across treatment sessions. The total intervention duration was eight weeks, with a treatment frequency of two sessions per week. This was a cumulative treatment intensity of 4,320 (Dose (270) × Dose Frequency (2x per week) × Intervention Duration (8 weeks)). CP’s treatment was similar, but he received 67 or 68 doses for four words per session, which equaled a total of 270 doses per session, which results in an identical cumulative treatment intensity to LM (e.g., 4,320). All other treatment parameters were equivalent to LM’s.

In order to ensure that the control words had the possibility of being produced by the child, clinicians were required to interact with examples of the control words, and talk about them without actually saying the control word. So, while playing with a car (target) and truck (control), a clinician would interact with both objects and say something like, “I like to play with the car. This one is fun too.” (referring to the truck). For CP, each treatment session included one elicitation (e.g. “What is this?”) of each word, followed by an expectant pause that created an opportunity for CP to produce target and control words, but did not model the word, or require him to speak. Research assistants in charge of fidelity ensured that examples of the control items were presented and interacted with during the sessions.

Each child had his mother present in the room during all treatment sessions. The main idea of the study was explained to the parents, but they were never explicitly told specifics such as dose number or what the target and control words were. Parents were welcome to be present during the sessions, but were asked to not participate in the session verbally if possible. If a parent happened to say either a target or control word, it was counted as a dose.

Fidelity and Reliability

In order to ensure fidelity of the treatment, research clinicians were trained on the VAULT procedure and were overseen by certified speech-language pathologists throughout the course of treatment. During each treatment session a research assistant, who had demonstrated accuracy both via video practice sessions and live treatment sessions, tallied each time a target or control word was said. She would periodically let the clinician know how many doses remained for each target, so the clinician could pace herself. The research assistant would refer to each target by the first letter of the word, so as not to provide a decontextualized dose. She would also alert the clinician if she happened to accidentally say one of the control words. Fidelity was calculated for each session, and was high, with an average of 97.36% adherence to protocol (range 92.95% – 99.27%), as measured by difference from the prescribed dose for each target word. Control words were rarely produced by the clinician (range 0 – 27), with ‘help’ being the most commonly produced control word with 27 examples across 16 sessions.

Outcome measures were: 1) the child’s spontaneous productions of the target/control words during the sessions, 2) the child’s spontaneous production of novel words that were not target/control words during the sessions, 3) the child’s production of the target/control words during a post-test probe immediately following treatment,4) the child’s production of the target/control words during a post-test probe 4 weeks and 4 to 5 days following treatment, and 5) the child’s performance on the MCDI at both post-treatment probes.

A separate researcher was present during each session to ensure reliability of the production scoring. When the child would produce a word, the clinician wrote down the child’s utterance and verbally repeated the child’s utterance so that she and the reliability checker could have verbal confirmation of the interpretation of the utterance. The parent was included in this step as well to ensure that interpretations of phonologically-immature forms were agreed upon by all parties. A phonologically immature production would be counted as a true production if it was bolstered by the context. For example, if a child made an approximation of ‘book’ while holding or pointing at a book, it would be counted as word. Alternatively, if the child made an approximation that might sound like ‘book’, but there was no context for the child to use that word, it would not be counted as a word. At the end of each session, the clinician and the reliability checker would confer on what they had heard, focusing on unique words, not repetitions. For LM, the average reliability was 92%, with a range of 33% −100%. For CP, reliability averaged 99%, with a range of 86–100%. If there were discrepancies, a third listener would watch a videotape of the session, and make a judgment about the utterance. Only utterances that were clearly interpretable were included in the data.

Results: VAULT

LM:

At his baseline trials before treatment began, LM produced none of the target or control words. Throughout the eight weeks of the VAULT protocol, LM produced two of the target words during sessions: “eat” once and “bye” three times, and none of the control words. Data for these sessions are in Figure 2. We calculated a treatment effect size for target and control words by averaging the number of words said in the last three sessions and divided by the standard deviation of those same sessions (e.g., Alt et al., 2020; Plante et al., 2014). The effect size was d = 0.57 for target words (i.e., medium); we could not calculate an effect size for control words, as there were no productions. Immediately post treatment, LM produced no target or control words during probes, although during the follow-up session two months after completing treatment, he produced one target word and one control word. In all, both the family and the research team did not feel that LM had made clinically significant gains.

Figure 2:

The number of target words LM produced during each session during the VAULT protocol.

Given that the VAULT protocol is designed on principles of statistical learning, we expected to see growth in vocabulary for words besides those that were trained. That is, we expect to see children in VAULT ‘learn to learn’ (e.g., Alt et al., 2014). At his baseline trials before treatment began, LM’s mother reported his use of 15 words on the MCDI, most of which were animal or environmental sounds and people or routines. His performance on the MCDI placed him below the 5th percentile for expressive vocabulary compared to children his age. Following the two-month delay period prior to treatment, LM was reported to have produced 22 words on the MCDI. Of these, eight were animal or environmental sounds, four were words for people, and five were games or routines. That is, 17 of the 22 words on the MCDI that LM produced were either linguistically-meaningful sounds or highly routinized utterances. He gained 0.9 words per week during the delay period. Following treatment, he was reported to produce a total of 25 words on the MCDI, gaining new words at a rate of 0.29 words per week. During the follow-up session two months after the end of treatment, LM was reported to produce 45 words on the MCDI, gaining new words at a rate of 3.89 words per week. Throughout the delay, treatment, and follow-up periods, LM remained below the 5th percentile for expressive vocabulary compared to his age-matched peers. Thus, he did not demonstrate clear evidence of ‘learning to learn.’

CP:

At baseline, CP produced none of the target or control words, and used 10 words on the MCDI, placing him below the 5^th percentile for his age. During the VAULT treatment, he produced the target word “ball” twice during a single session. He did not produce any of the control words. Data for these sessions are found in Figure 3. We could not calculate a treatment effect size, as there were no productions in the last three sessions. CP produced one target word during a probe session immediately following treatment, but no target or control words during a follow-up four weeks after VAULT. Post VAULT, CP gained 8 words on the MCDI, at a rate of 0.76 new words per week of treatment. He gained no words during the four weeks following VAULT. CP produced 8 novel words during VAULT treatment, at a rate of 1.12 words per week. Thus, at the end of VAULT he remained below the 5^th percentile on the MCDI, did not regularly use any of the treatment words, and showed no evidence of ‘learning to learn’.

Figure 3.

The number of target words CP produced during each session during the VAULT protocol.

Qualitative Description Performance during VAULT

While the input-based intervention was unsuccessful in efficiently supporting LM and CP’s spoken language development, both children consistently demonstrated intent to communicate throughout duration of this treatment protocol. This was evidenced by their engagement and participation in child-led therapy activities. For example, LM consistently greeted the clinicians with an enthusiastic social smile in the waiting room. LM was often excited to begin a treatment session, grabbing clinicians by the hand to lead them into the therapy rooms. Because both children had such limited word usage, they communicated in alternative ways. They had a heavy reliance on forms of prelinguistic communication, consistent with patterns of gestural emergence commonly identified in late-talkers (Thal & Tobias, 1992).

Overall, both LM and CP were very successful nonlinguistic communicators. They demonstrated intentional communication skills, making efforts to gain the attention of the clinician (e.g., via eye contact, wait time, vocalization) coupled with an effort to communicate (e.g., via vocalization and/or gesture or occasionally via non-linguistic verbalization). LM and CP maintained eye contact with clinicians during social and verbal interactions. They both engaged in deictic gaze and were attentive to shared topics (e.g., books, toys, images) in communicative interactions. They were able to maintain joint attention and joint action during sessions. LM consistently engaged in sharing and regularly invited the clinician to participate in collaborative play using symbolic gesture and pre-linguistic vocalizations. Both children used their protowords (/ma/ for LM and /da/ for CP) for many communicative functions, including to make requests, communicate desires or preferences, indicate closure, and fill communicative turns, among other purposes. LM used both deictic gestures (e.g., pointing, showing) and symbolic gestures (e.g., pushing a toy car or pretending to eat play food). He used gestures alone and in combination with prelinguistic vocalizations, such as grunts or vowel sounds. Both children filled the majority of their communicative turns and their responses were timely and appropriate. Of note, both children were able to convey complex messages without the use of words. For example, LM was able to use gesture, vocalization, body language, and wait time to indicate that he wanted to race down the hallway at the end of a session. CP communicated the message, “I want bubbles so the dinosaurs can eat them.” nonverbally by pointing, acting out the dinosaurs eating, and using a self-made sign for bubbles.

Overall, both children demonstrated an understanding, willingness to, and ability to communicate meaningfully, but not to communicate meaningfully with a variety of spoken words. We felt the pragmatic concept of meaningful and specific word production needed to be taught in order to encourage increased reliance on verbal rather than nonverbal communication.

Methods: Trial Therapy

The trial therapy was also approved by an IRB at the University of Arizona. This component of the treatment was exploratory. We sought to teach these boys the importance of using conventional lexical labels to communicate and to give them a way to easily and successfully produce specific lexical labels. We considered that their negligible response to the VAULT protocol may have been due to a lack of pragmatic awareness of the requirement for meaningful word use because they heavily relied on nonverbal communication. Given that they were so often successful with prelinguistic communication, they were not making an efficient transition to including regular, specific, and varied word use into their repertoires of communication skills. The input-based nature of the VAULT protocol allowed these boys to successfully communicate during treatment using only their protowords (e.g., /ma/, /da/) or non-linguistic communication (e.g., gestures, gaze, pointing). We suspected that the VAULT protocol failed to elicit vocabulary growth in these particular toddlers because it relied on them to spontaneously produce meaningful, unique words following sufficient input.

It was also possible that the boys’ speech sound delays contributed to their lack of linguistic output. LM’s and CP’s speech were characterized by a reduced phonetic inventory and syllable structure, as is common for late talkers (Mirak & Rescorla, 1998; Paul & Jennings, 1992). As such, we thought the use of an AAC device might provide the boys a quicker route for their output to match the conventional phonology of words. That is, using an AAC would allow them to immediately produce a word others could easily understand. Given their strong receptive vocabulary, it is likely that both boys had the capability to produce their target vocabulary words if they were supplied with the adequate tools, environment, and reinforcement.

We developed a trial therapy framework with the intention of addressing both of LM’s deficiencies, which we then extended to CP, as he showed characteristics similar to LM. Use of low-tech AAC has been shown to increase speech output in a variety of language-delayed children, although consistent with most of the literature, the children reported in these studies tended to have intellectual disabilities or autism (Millar, Light, & Schlosser, 2006). Furthermore, late-talkers with speech sound delays, including a small phonetic inventory, may be more responsive to an AAC environment, since they can bypass the execution and planning demands of speech while having a successful communication interaction. As a result, having an alternative means to communicate when speech fails may make attempting speech more likely (Leech & Cress, 2011).

We addressed the pragmatic component of each toddlers’ deficiencies by developing a therapy framework that required them to make specific requests in order to achieve desired objects or tasks. Although we did not make the children say the words using their voices, we did require them to produce words using the AAC device in order to obtain the desired object/activity. Part of this was motivated by Vlach and Sandhofer’s (2012) finding that learners improved their word learning retention when they were prompted to say target words, in direct contrast to VAULT, which focused solely on providing input. While the VAULT protocol was designed to be input-based and reduce pressure on children to comply with requests to imitate words, evidence from the fast-mapping literature shows that producing a word may help a learner retain it (Vlach & Sandhofer, 2012).

Trial Therapy Procedures: LM

Trial Therapy Materials.

The clinicians presented LM with a variety of toys and objects of interest, such as plastic jumping frogs, a toy castle, a male action figure, and plastic eggs filled with pictures or toys. He was also presented with a single-message, switch-activated speech generating device (i.e., a BIGmack talker; Ablenet, n.d.) prepared with the word “more.” Following exposure to the single-message speech generating device, LM was presented with a 3-choice, switch-activated speech generating device, containing the familiar word “more” as well as other relevant words available in the play context. These included: “frog,” “egg,” “castle,” and “guy” (referring to the action figure). Pictures of the referents were taped onto the corresponding buttons.

Trial Therapy Design.

The treatment protocol remained similar to the original VAULT procedure in the following ways: the sessions were conducted in the same treatment room by a graduate clinician supervised by a certified speech-language pathologist. The clinician narrated the child’s play. All spoken or AAC-generated utterances were repeated immediately by the clinician. The speech-language pathologist recorded unique spoken utterances for proof of concept. LM’s four sessions were distributed across two weeks, with two sessions per week.

The VAULT protocol was further adapted to meet LM’s needs. The focus of the sessions shifted from input-based with targeted vocabulary to a flexible, contingent response approach. In this case, the goal was to elicit any language output, not necessarily vocal speech. For this reason, all linguistic output (i.e., speech or symbol selection) was reinforced with praise and presentation of the requested item or action (i.e., “help”). The clinician presented LM with toys of interest and modeled use of the single-message AAC device to request “more,” following best practices for AAC use (e.g., Sennott, Light, & McNaughton, 2016). She then withheld objects of interest until LM requested them by either using the AAC device or attempting to produce a word. As LM demonstrated awareness of the purpose for the single-message device (i.e., using the single messages in a pragmatically appropriate way such as naming an object, or requesting an object), the 3-choice, switch-activated device was presented containing relevant words for the available toys and activities. When LM demonstrated emerging competency activating single words on the 3-choice speech device by using specific words in a meaningful way (i.e., naming/requesting), the device was adapted to contain phrases such as “I want” to encourage combination of symbols during play.

Trial Therapy Procedures: CP

Trial Therapy Materials.

The clinicians continued to use toys and activities that targeted CP’s VAULT words. He was also presented with several single-message, switch-activated speech generating devices (i.e., a BIGmack talker; Ablenet, n.d.) prepared with his four target words: “truck” “arm” “shirt” and “hungry”. Pictures of the referents were taped onto the corresponding buttons. We began by introducing one button at a time. After CP showed an understanding of how to use the button, others were introduced, so that he could choose between one, two, three, or four buttons at a time.

Trial Therapy Design.

The design was similar to LM’s sessions, except that we retained a focus on the target words from VAULT, although the primary goal of the session was to encourage output, either vocal or AAC-generated, using a flexible, contingent response approach, as described in LM’s section. For CP, we included the expectant pause (“What is this?”) included in his version of VAULT as well as some obligatory prompts (e.g., “Tell me which one you want.”). Recall that expectant pauses did not require a response, but for obligatory prompts, the clinician withheld objects in order to encourage CP to respond. CP was seen for three trial therapy sessions, distributed one time per week for three weeks.

Results: Trial Therapy

LM:

Because the AAC protocol was dynamic and informal, data collection of LM’s utterances was not exhaustive. Due to the experimental nature of the trial therapy and LM’s exponential improvement, our data provide a low representation of LM’s utterances and productions because we discontinued documenting individual productions following multiple repeated utterances; we were satisfied to count the number of different words¹

Within the first session of AAC therapy, LM vocally produced at least 44 words, with 17 unique words. Combining data from LM’s speech and his device output, LM produced at least46 words. During his second session, he vocally produced at least 73 words, with 24 unique words. During his final AAC trial session, LM vocally produced at least 79 words, with 22 unique words. Throughout the four trial sessions with AAC supports, he vocally produced over 196 spoken words. (Note: data were only available for sessions 1, 2, and 4.) When including words produced with multimodalities of speech and the AAC device, he produced over 224 words in four sessions. Of these, 23 were novel words that were not reported on any of his MCDI forms, nor had they been heard during VAULT treatment. During VAULT, we also counted the total number of words LM said, including words other than target or control words. This would be a clear comparison to our count during the AAC therapy. During VAULT, LM produced an average of 5.25 words per session, with a range of 0–16. His average use of unique words per session during VAULT was 2.31 (range 0–7). Thus, his productions during the trial treatment were 13 times better for total words produced, and 9 times better for unique words produced.

Additionally, LM produced his first multiword combinations during the trial therapy. He began combining vocally produced words into two-word combinations and attempted vocal production of a four-word phrase during his final session. Although he did not combine words during his first session with AAC supports, during the second session, LM combined AAC-produced and vocally-produced words into 10 multiword phrases (e.g., “I want frog” was produced with “I want” via the AAC device and “frog” spoken). During his final session, he produced 18 multiword utterances, six of which were produced with multiple modalities of speech and AAC, and 12 of which were produced via speech only. This compares to zero multiword productions during VAULT.

CP:

During his first trial therapy session, CP vocally produced an approximation of his target word “hungry”. He produced AAC-generated responses both spontaneously and in response to expectant pauses for all four of his target words. His use of the BIGMack was not perfect in that he exhibited several non-responses to obligatory prompts and made several unintended button presses as well. However, over the course of the 30-minute session, he produced at least 16 accurate productions of his target words and began to differentiate between buttons, accurately selecting from fields of three and four.

During his second trial therapy session, CP vocally produced an approximation of his target word “arm”. He produced an AAC-generated production of all four target words throughout the treatment, both spontaneously, as well as following expectant pauses and sometimes following the clinician’s model. For “truck”, “shirt” and “arm”, he produced spontaneous AAC-generated productions (at least 22 combined) and for “hungry”, he produced four AAC-generated productions in response to an obligated prompt. He also combined an AAC-production with pointing to an object. Twice during the session he chose the incorrect button for his intended meaning. At times, he experimented with the AAC devices, clearly playing with the device (e.g., multiple presses in a row in a solitary play scenario). We did not count these explorations as productive communicative attempts.

During the third session, CP did not produce any vocal productions of his target words. However, he accurately produced both spontaneous and elicited productions of all four target words using his AAC (at least 22). He again pointed to an object in conjunction with an AAC-generated production (i.e., pointed to a crocodile while pressing the “hungry” AAC button in a play scenario in which he was feeding the crocodile). He also clearly demonstrated understanding of the communicative function of the AAC. While playing with a toy that we thought was unrelated, he pressed the “hungry” button. We responded with confusion, but CP persisted and showed us that they toy he was playing with had a band-aid with Cookie Monster on it, and the “hungry” was referencing Cookie Monster. CP continued to experiment with the AAC devices, for example, pressing all the buttons at once, but would then return to using them in a communicative manner. He used nonverbal cues (e.g., eye contact, which way his body was facing) to signal when he was exploring or communicating.

Discussion

The incorporation of AAC into treatment yielded positive results for these two toddlers. This supports the findings from the literature that show positive communication outcomes for AAC use for toddlers (e.g., Romski et al., 2015; Solomon-Rice & Soto, 2014). However, it is one of the first studies in the literature to explicitly and intentionally include children with LLE with an AAC protocol. Solomon-Rice and Soto’s study included children with global developmental delays or histories of prematurity who were already using multi-modal, albeit low-tech AAC systems. For two of the three participants in their study who showed positive outcomes, the AAC condition in which the clinician modeled use of the AAC device was not necessary for their success. That is, they both learned equally well in a focused-stimulation condition that did not model AAC use. However, they had AAC systems in place; the difference was in the clinician input.

In contrast, our toddlers were not successful in a focused-stimulation setting that did not include AAC. Although input-based therapy is effective for some late-talkers (Alt et al., 2014; Alt et al, 2020), it did not result in the desired outcomes for LM or CP. These children required a modification to the approach in order to improve their expressive language. This is in line with the findings from Romski et al. (2010) who found that toddlers (albeit with more profound disabilities than the toddlers in our study) produced more spoken words when trained using AAC than when trained only with spoken communication. Upon the transition from the VAULT protocol to the secondary, exploratory method, LM showed immediate, significant improvement in his number of spoken productions, use of novel words, and use of multiword utterances, demonstrating an emergent syntactic skill of combining words. The magnitude of his improvement was striking to both the clinicians and his primary caregiver. CP’s improvement, while not as dramatic, was still notable. Both children increased the number of words spoken during treatment by over 600%. The implications here are that, for some children who are late-talkers, the addition of an AAC device to an input based treatment may be necessary for success. Additionally, if this is the child’s first introduction to AAC (which is likely, given the limited use of this tool for children who are late-talkers), there may need to be an adjustment to the input-based protocol that puts more demands on a child for output. This serves the dual purpose of teaching the child how to use the AAC device as well as highlighting the pragmatic need to contribute to conversation, which is a challenge we outline below.

Toddlers’ challenges

Both children shared similar challenges. Prior to the start of trial therapy, both children successfully communicated through nonverbal modalities, following the pattern outlined in the literature by Thal and colleagues (1991;1992). For LM this was mainly through gestures, gaze, protowords, and body language. CP communicated primarily through sound effects, pointing, and acting things out. Given the high success of their prelinguistic communication, the authors suspect that these boys were not incentivized enough to develop specific word use. While non-linguistic communication served the boys for the demands of their present age and environment, use of only these modalities would have likely become problematic as they aged. For example, LM sometimes communicated through brute force (e.g., yanking on a door, pushing) which could be dangerous as he gained in strength. As a child grows, the social and functional demands of his life will increase. Individuals use specific means of communication to have success addressing their wants, needs, feelings, ideas, and social etiquette. Additionally, as these boys enter school, they will require means of communication that support academic success as well as creating and maintaining peer and adult relationships. Importantly, LM’s mother reported that LM was becoming increasingly frustrated with his communication attempts. CP’s parents reported less frustration (both for the family, and the child) than LM’s family, but CP’s family was still concerned about his lack of speech.

We also suspected that both children’s speech sound delays and limited phonetic inventory may have contributed to their low expressive output. For example, many of LM’s attempts to vocalize his requests resulted in highly unintelligible or unclear productions. This aligned with our perception of a speech sound delay. As an example of limited phonetic inventory, instead of saying ‘yes’ or ‘no’, CP used the pragmatically appropriate, but phonemically less demanding ‘mmhmm’ or ‘uhuh’. This presentation of limited phonetic inventory mirrors what the literature reports for late talkers (e.g., Mirak & Rescorla, 1998; Paul & Jennings, 1992) and is a concern, given the findings that toddlers with difficulty with speech sound production challenges are likely to have longer lasting problems than peers without speech challenges (MacRoy-Higgins & Schwartz, 2013).

Finally, both children’s continued use of a protoword suggested that they were not grasping the ‘pragmatics’ of using oral language, that is, that each word needs to sound distinctly different from other words As a result, introducing an AAC system reduced the burden of spoken productions, while providing an alternate means to demonstrate their expressive language capability. High rate of protoword use is another characteristic of late talkers that has been reported in the literature (Hodges et al., 2017).

The influence of AAC on verbal communication

We trialed a BIGmack talker as a “quick and easy” means to allow these boys to circumvent their difficulty with speech sound production by using the gross motor skills of their hands rather than the fine motor skills of the speech mechanism in order to make requests. After giving both boys a chance to explore the AAC system, the protocol was altered to require a production before the children receiving their requests. This change to the VAULT protocol allowed the boys to experience successful communication attempts, supporting the pragmatic purpose of specific word use beyond their nonverbal communication. The use of the AAC device allowed them to recognize the effectiveness and efficiency of using contextualized, conventional vocabulary.

With these adjustments to the treatment protocol, these children had two communication means for meeting the new demands of requesting before reward: through the AAC device and speech. The single-button talker allowed LM to articulate his specific requests while bypassing the planning and execution demands for verbal speech. That said, verbal speech has greater flexibility and portability than an AAC device, and for that reason, it is the more efficient means of communication. Once LM experienced successful communication with the support of an AAC device, he chose to attempt vocal productions to complete his requests. As similarly demonstrated in the literature (Leech & Cress, 2011), use of AAC in our study did not prevent speech production or delay speech development. Instead, the use of AAC supported LM’s individual communication needs, which in turn supported his speech production. CP also made more vocal attempts at his target words during the treatment sessions that included AAC, but most of his productions were made using the AAC.

We used a BIGmack switch because it was available, but any low-tech AAC strategy would likely have resulted in similar outcomes. The trialed AAC system was used as a way for these children to communicate specific words without requiring the accurate fine motor skill to produce the words vocally. Other AAC systems, for example low-tech boards or picture exchange systems could have similarly circumvented a speech sound delay. Once LM was provided with a treatment approach that scaffolded his individual communication needs, he showed significant and rapid improvement of his expressive language, not only with single-word requests, but with use of multi-word phrases. By combining multiple words, LM demonstrated rapid improvement to his use of expressive linguistic complexity, a skill that had not yet emerged prior to trial treatment. CP’s improvement was not as dramatic, but he, too, made impressive gains in the use of conventional words used in appropriate communication attempts via the AAC device.

Clinical Implications

The present clients’ demonstrated success following the change in protocol leads us to three major implications for when input-based early intervention fails. First, prior to success in input-based treatment, a child may need targeted therapy to develop the pragmatic skills for intentional, specific word use. Second, an AAC device may provide a means for teaching this skill while allowing for circumvention of intelligible speech production demands, even for toddlers whose primary diagnosis is LLE, who are typically excluded from research on AAC use. Although we used a BIGMack switch, there are many types of speech generating devices a clinician might choose (e.g., tablets), that can be quickly trialed without the need for an extensive AAC evaluation. The finding that children with LLE are at risk for long-term speech accuracy issues (Neam et al., 2019) heightens the need to focus on speech skills in toddlers with LLE. Finally, increasing output demands for a child during early intervention therapy may increase their attempts at spoken communication once they have experienced successful communication. There may need to be a balance between high-demand output-focused therapy and low-demand input focused therapy. When making this clinical decision, one might pay attention to the characteristics of the child. We selected these two toddlers because it seemed like they needed a better understanding of the need to use language in a conventional way, rather than using protowords.

Children with large gaps between their expressive and receptive language skills are likely candidates for this type of therapy, particularly when they unexpectedly do not respond to evidence-based input-based early intervention. Additionally, children exhibiting speech sound delays will likely need further therapy targeting articulation, phonology, or both, following increased number of vocal utterances. AAC therapy provides an option for scaffolding expressive linguistic output in this population before targeting articulation or phonological skills.

An additional consideration is the importance of reporting modifications to treatment protocols in the literature. This type of reporting is required when using standard reporting guidelines like the Template for Intervention Description and Replication (TIDieR; Campbell et al., 2018), which includes sections to list modifications. In treatment research, and thus, in clinical practice, it is not uncommon to have non-responders or participants who do not make the expected gains that others do. Thus, clinicians must often think about why a person might not respond, and make modifications to a treatment approach in an attempt to best meet a non-responder’s needs. The thought process and outcomes of these attempts should be documented so others can attempt to replicate the approach, and build an evidence base for the modification.

Limitations

Single-subject studies such as these have inherent limitations which are worth addressing. Because this study looked at two participants, we cannot readily generalize the reported results to a larger population. Our trial therapy was intentionally brief for both toddlers. We wanted to test a concept, with the hope of providing the children’s future SLP with some techniques that could be fruitful. Clearly, these few sessions are not representative of a full course of treatment. In addition, due to the exploratory nature of the trial therapy, there were two major changes to the VAULT therapy protocol that were implemented simultaneously (i.e., addition of an AAC device and increasing production demands). As a result, neither variable can be isolated as being solely responsible for the change observed in the boys’ expressive language. For example, it is possible that adding in demands for production (without AAC) would have led to the same type of outcome. While we feel both variables contributed to the toddlers’ improvement (particularly because their parents at times made demands for productions that went unanswered, or made infrequent demands), future research may wish to target each of these variables in isolation. Additionally, both children followed the same order of treatment: VAULT, then AAC. There is the possibility that a treatment like VAULT may have been successful following a treatment that established AAC as a means for expressive vocabulary production. These issues of timing need to be resolved in future studies.

During trial therapy, clinicians stopped collecting data on LM’s number of utterances after proof of concept was established; for that reason, he may have produced more spoken utterances than reported here. Alternatively, one might have noticed that LM’s rate of word learning on the MCDI in the follow-up measurement from VAULT was a relatively high (for him) 4.2 words per week, and thus surmise that the improvement we saw was simply maturation. However, although we did not collect a final MCDI, LM’s use of novel words during the 1.28 weeks of the trial treatment resulted in a rate of 17.96 new words per week, which was much higher than his previous rates of growth, making maturation an unlikely explanation. We did not collect a final MCDI for CP. He primarily used the AAC device to communicate, so did not verbally produce new words outside of the target words. Finally, both the VAULT study and the trial therapy controlled for a clinician effect by having two graduate student speech-language pathologists conduct therapy. A limitation, however, is that the VAULT study and the trial therapy were conducted by different pairs of clinicians. For that reason, it is possible that the change in clinicians may have also had an impact on the boys’ improvement. However, their improvement was so dramatic, and given that most literature on AAC use with children their age excludes children with their profile (i.e., LLE, a primary speech/language diagnosis), we felt it was worthwhile to share these exploratory case studies with the field.

Future Research

Ideally, future research will investigate the best methods to establish the concept of meaningful word production in children. This study shows that the use of AAC is a viable choice for children who are often not considered for AAC use due to either age (e.g., Cress & Marvin, 2003) or due to a less severe diagnosis than is typical for AAC use. With children like this, clinicians might consider using AAC as a treatment tool, without needing to initiate the important but time-consuming protocol for a dedicated AAC system. For children with less severe challenges, it may not always be necessary to undertake a formal AAC evaluation, feature matching, and funding in order to trial AAC in therapy. We urge clinicians to view AAC as a stepping-stone available to them to help in the development of language skills in their clients whose primary diagnosis is LLE. This is in line with Romski and Sevcik’s (2005) idea that AAC serves many purposes in intervention and can be a temporary strategy. Further research may examine whether facilitating linguistic output opportunities (both spoken and AAC) in a mainly input-model approach improves linguistic output in early intervention. Additionally, it would be ideal to study the adjusted trial therapy approach with a greater number of children and with children with other speech and language characteristics. This research may provide evidence to support including AAC in a variety of therapies when other evidence-based methods fail, or as an initial approach with children.

Highlights:

• Demonstrated the outcome of the use of AAC with two late-talking toddlers relative to vocabulary production

• Highlighted characteristics of the toddlers that may have made them more receptive to AAC use.

• Suggested that clinicians and researchers rethink the most common recipients of AAC use, and attend to the existing, but limited, literature/guidance documents that advocate for the use of AAC for children without severe disabilities.