A substantial literature on children with Down syndrome (DS) suggests that deficits in language acquisition are greater than deficits in some other aspects of development such as nonverbal cognitive skills. Deficits in both vocabulary and morphosyntax have been identified in expressive and receptive modalities, although receptive language skills are reported to be generally stronger than expressive language skills [1-3]. These deficits are particularly apparent in studies where children with DS are compared to non- DS peers with intellectual disabilities matched on mental age or IQ level [4-7]. Because vocabulary acquisition is a crucial element of language development in all children, but especially in those with delayed or inefficient language learning, additional research on expressive and receptive vocabulary intervention, and cross modal generalization between these modalities in DS is warranted [8].

During language intervention, clinicians can potentially target either expressive and/or eceptive skills because ac child with DS is highly likely to display significant deficits in both modalities. For example, clinicians could ask children to name objects and/or point to pictures from a stimuli set or a storybook page. However, these activities may not be systematically coordinated or integrated in clinical practice and expressive and receptive vocabulary interventions have not often been studied simultaneously or in combination in populations with intellectual disabilities, including DS [9]. A number of studies have considered the effects of expressive-only language intervention on expressive and receptive language outcomes in populations with language deficits [10]. And a few studies have considered the effects of sequential expressive and receptive language interventions, including vocabulary interventions [11].

However, expressive and receptive language modalities have not been studied to explicitly examine cross modal generalization in DS, wherein each modality is trained apart from the other on separate, independent ocabulary targets. Because receptive language is strongly associated with long term language outcomes and with subsequent literacy skills in populations with typical and atypical language, including DS [12,13], there is an overarching need for systematic study of receptive language intervention on outcome. Of course, if children are able to make gains in receptive language as a secondary or incidental effect of expressiveonly language intervention, then expressive language intervention exclusively may be an efficient program for training words. However, a study by Camarata et al. [10] demonstrated that children with poor receptive morphosyntax (SLI-expressive/receptive) at the onset of therapy did not make progress through expressive-only language intervention. These findings suggest that receptive language must be explicitly taught for some populations with poor receptive language skills. Studies of children with DS or unspecified ID have largely found little to no evidence of generalization, but there are few studies that examine generalization in vocabulary [14-17].

“Cross modal generalization” is defined herein as when a target word is taught in one modality (e.g., expressive) and accurately identified via another modality without direct instruction in that modality (e.g., receptive). Explicit training is delivered in only one modality to test cross modal generalization. For example, after the word “clock” is trained receptively, the word “clock” is subsequently produced when a picture of a cowboy is presented (without direct instruction on naming). Or, an example of expressive to receptive generalization is seen when the word “cowboy” is taught expressively, and then “cowboy” is correctly selected from an array by pointing when requested. Typically developing children show high levels of cross modal generalization at the word level, receptive-to-expressive as well as expressive-to-receptive [18-20]. However, children with SLI [20] and children with DS matched on MA [18] were not as successful as their typically developing peers. Children with intellectual disabilities do not show a similar propensity for bidirectional cross modal generalization at the level of morphosyntax [14,15,17].

To date, there is a striking paucity of information on cross modal generalization in children with DS. An early study of cross modal generalization included two children with DS who unsuccessfully generalized morphological information regardless of taught modality [15]. Studies of morphological generalization (for example, from comparative forms such as “bigger” to superlative forms such as “biggest”) in children with DS and other intellectual disabilities have also demonstrated relatively high levels of target learning, but relatively low levels of generalization [14-17]. Although generalization of morphology appears consistently low in populations with DS, a comprehensive literature identified three studies examining the effects of cross-modal generalization at the word level in children with intellectual disabilities, none included children with DS. Anderson and Spradlin [21] and Bucher and Keller [22] examined unidirectional cross modal generalization at the word level with variable results across different children. Keller and Bucher [23] was published as an abstract, so detailed findings are unavailable, but the authors reported that receptive-only intervention was insufficient for cross modal generalization in children with severe language deficits.

Given the inherent clinical advantage of training targets that readily generalize, and given the heterogeneity across disability typologies and also across individuals within the same typology, systematic study of cross modal generalization in children with DS is warranted. Studying cross-modal generalization may have implications for efficient word learning strategies and the design of effective lexical or vocabulary interventions in children with DS. The purpose of this study was to examine whether children with DS generalize learned vocabulary, specifically nouns, from expressive-to-receptive and from receptive-to-expressive modalities. Two research questions were considered in the present study: (a) Do the children with DS generalize cross-modally from expressively-trained targets to receptive target identification? And (b) Do the children with DS generalize cross-modally from receptivelytrained targets to expressive target identification?

Study design

A multiple probe single case design [24,25] was employed to examine cross modal generalization of targeted vocabulary words. The design consisted of concurrent multiple probe designs and included six sets of intra-subject replication (three expressive and three receptive sets). The project was reviewed and approved by the Vanderbilt Institutional Review Board (IRB) for the protection of Human Subjects and consent was obtained from the parents of all participants prior to enrollment in the study.

Five children with DS were taught vocabulary words that were absent in baseline expressive probes (0%) and at or below chance levels (<25%) in receptive probes. Intervention modality was randomly assigned at the word level so that half of the words were trained expressively and half of the words were trained receptively. Individual words were trained in one modality: Targets were trained in either receptive or expressive modality exclusively; no cross modality training was delivered. Identification of vocabulary words was measured at baseline probes, during the intervention condition, and during follow-up probe conditions. Probe conditions at baseline and follow-up included all targets vocabulary words, but probes during intervention included only targeted words.

Participants

Participants included five preschool children with Down syndrome, ages 3 to 5 years with a mean age of 4; 6 (years; months). The children (four males, one female; assigned pseudonyms below) were recruited through the Bill Wilkerson Center of Vanderbilt University and from Nashville Metro Public Schools. Participants were identified through parent and clinician report as qualifying for language intervention services and validated by subsequent standardized testing. All children who met eligibility requirements were included in the study. Inclusionary criteria were: (a) a diagnosis of Down syndrome through physician report, (b) ability to imitate vocally, and (c) a commitment to enrollment in intervention for 6-8 months. The pseudonym for the female participant herein is “Joy” and the pseudonyms for the males are Sam, Nick, George, and Tony.

Children’s intellectual abilities were measured using the Revised Leiter International Performance Scale [26]. Participants also received preintervention testing using the Test of Auditory Language Comprehension-3 (TACL-3) [27], a receptive measure that includes grammar and vocabulary test items; the Preschool Language Scale – 3 (PLS-3) [28], an omnibus language test, the Peabody Picture Vocabulary Test -4 (PPVT-4) [29], a receptive vocabulary test, and the Expressive One-Word Picture Vocabulary Test - 4 (EOWPVT-4) [30], an expressive vocabulary test. Of the five children, two children demonstrated receptive scores above expressive knowledge (Sam and Joy), one child demonstrated receptive scores equal to expressive (Nick), and one child demonstrated receptive scores lower than expressive (Tony), although numerical differences in receptive and expressive standard scores fell within each test’s 95% confidence interval. The final child, George, had no correct responses on either receptive or expressive vocabulary measures prior to initiating the intervention procedures. These results are summarized in Table 1. Children are listed in the order in which they enrolled in the study.

Table 1: Description of participants and standard test results.

Vocabulary selection

A minimum of 16 vocabulary words were identified for each participant. A five-step process was employed to identify a set of target vocabulary words, individualized for each child, that were absent from the child’s expressive and receptive vocabulary repertoire. First, a pool of words was selected and each child’s parent indicated which words were absent from their child’s repertoire. Second, the experimenter engaged each child in a word imitation task, to identify those words each child was unable to produce an intelligible pronunciation; the goal of this task was to ensure that a child’s phonological limitations would not confound expressive generalization. Third, the child was asked to match one picture for each of the vocabulary words to an identical picture. The clinician did not name the vocabulary word; the child was asked simply to match the pictures. Only those stimuli that the child could correctly match were included. This task was included to ensure that the child had the necessary prerequisite matching skills for each potential target so that receptive probes could be completed accurately and that errors in this modality could not be attributed to an inability to match or point to a target form.

Fourth, the child was asked to identify each word expressively and receptively. To test for expressive knowledge, the child was asked to name each picture. Only pictures that were not named correctly were selected for intervention (0% accuracy across four opportunities). To test for receptive knowledge, the child was asked to identify each picture from a set of four choices. Only pictures that were identified at or below chance level (<25% correct across four opportunities) were selected for intervention. These procedures ensured that all targets were absent from each child’s expressive and receptive repertoire and that this absence was not due to an inability to match or point to a picture or due to an inability to intelligibly speak a target form. Finally, words were randomly assigned to a modality (expressive or receptive) and a condition (sets 1, 2, 3 or control). A sample process for one child in the study, Tony, is provided in Appendix A.

It is possible that incidental (untrained) cross modal learning could arise from employing real words in this study. So, a set of four untreated target nouns (two receptive and two expressive) served as a control set. Although these words were not taught during intervention, they were included in each probe session to control for incidental learning. Therefore six expressive words and six receptive words were taught per participant (twelve total), two expressive and two receptive words were untreated but monitored (four total) in each participant. Thus, a total of eight expressive words and eight receptive words were included in each probe session (sixteen total). Order of presentation of these probe sets was randomly selected for each participant and the pairs selected for the control condition was also randomly drawn. Baseline procedures included at least three sessions of probes on all sixteen words (targeted and control) to establish a stable data pattern before the intervention condition was implemented.

Language intervention procedures

Intervention sessions were conducted in 1:1 (clinician:child) therapy setting at the Vanderbilt Bill Wilkerson Center by licensed speech-language pathologists. Children participated in an average of four sessions per week. During intervention, the clinician introduced targeted vocabulary words in a storybook context and a structured play context (Appendix B). The combined time for storybook and structured play was 12 minutes for one modality (24 minutes total per session). This combination approach was adopted because the goal of intervention was to ensure that targets were learned, not to evaluate the relative contribution of each component.

During the storybook context, the vocabulary words were verbally presented by the clinician at least 10 times per session while the clinician and child viewed a picture book and the clinician presented a story. Immediately after the clinician said the vocabulary word during the story, the clinician prompted the child (a) in the expressive condition, to name the vocabulary word when shown a picture that matched the word or (b) in the receptive condition, to identify the vocabulary word by pointing to the picture that matched the word (receptive condition). A play based vocabulary intervention followed the storybook context. During the play context, the clinician and child interacted with a set of toys that matched the storybook theme and included the target vocabulary items. The clinician commented on the play, using each target vocabulary word at least 10 times per session. For the first two children in the study (George and Nick), the clinician also prompted the child to name (expressive target words) or point to the items (receptive target words). For the other three children, the clinician named the targets while commenting, but did not deliver prompts to name or point. The clinician, play context, and reinforcers (see below) were identical across expressive and receptive conditions.

The clinician reinforced correct responses with verbal praise, smiles, and/or high-fives. For incorrect responses, the clinician delivered the correct response (“Uh-oh, it’s a_____________________ ” for expressive items or “uh-oh, here it is” and pointing to the correct picture for receptive items). If the child then identified or produced the word correctly, the clinician reinforced the correct response. If the child made no response after prompting, the clinician waited a few seconds and then delivered the correct expressive or receptive exemplar.

Probes

In total, three types of probes were used: (a) daily intervention probes that were administered at the outset of each intervention session in order to chart target learning, (b) within-modality probes that were administered during separate, dedicated probe sessions, and (c) cross-modal probes that were administered also during separate, dedicated probe sessions. All probes were administered by the same examiner who also administered the intervention to an individual child.

Daily intervention probes: At the outset of each intervention session, the child was asked to point to pictures of the target vocabulary words (for those words targeted in the receptive condition) and to name the targeted vocabulary (for those words targeted in the expressive condition). Only those words targeted for intervention in that treatment session were included in an individual daily intervention probe. For receptive probes, the child was asked to point to a targeted vocabulary word from an array of four pictures. For the first two children, George and Nick, the array included one target item and three distractors, picture that were not ever target words. But for the other three children, the array included one same-modality vocabulary word (not the correct response in the trial) and two distractors. These changes were made to test overgeneralization of within modality targets among each child’s selections. For expressive probes, the child was asked to name an individual picture. Non-corrective feedback (e.g. “nice pointing!”; “thank you for talking”) was given throughout the daily intervention probes. The order of probes delivered was randomized. After the child reached at least 80% correct responses across daily intervention probes for three consecutive sessions, the next probe condition was initiated as prelude to intervening on the next vocabulary set.

Cross modal generalization probes: At baseline and at each subsequent probe condition, the child was asked to point to pictures representing the targeted vocabulary words (expressive-to-receptive generalization probe) and to name the targeted vocabulary (receptive-to-expressive generalization probe). For example, if a child had learned or was scheduled to learn the word “cowboy” receptively, the child would be asked, “What’s that?” or prompted, “this is a___.” Cross modal probe data was collected once per probe condition, and at least three trials of each word were included each time vocabulary words were probed.

All words, including trained, untrained, and control sets of vocabulary, were included in the cross modal probe condition. Non-corrective feedback was used throughout the cross modal probe condition. In order to maintain motivation, “success words” were interspersed with the targeted vocabulary. These were words identified by the parent as well-known and motivating words for the child. Some examples of success words were favorite television characters including “Elmo” and “Barney.” These success words functioned to maintain a high level of success during the probe task and as a measure of ongoing task compliance.

Within-modality probes: During within-modality probes, each word was probed in the same modality that it would be (baseline) or had been (follow-up) taught. For example, if a child had learned, or was scheduled to learn, the word “cowboy” receptively, the child would be asked to “point to cowboy” as a within modality probe. Unlike daily intervention probes wherein session specific targets were probed, all words assigned to that participant, including trained, untrained, and control sets of vocabulary, were included in the within-modality probes. Non- corrective feedback and success items were used throughout the within modality probes.

Measuring generalization

A straightforward dependent measure of generalization was to plot percentage of correct responses for the receptive and expressive targets, and this method is included in the figures for each participant. In addition, it was also useful to adopt a more stringent criterion for generalization that could take into account the impact of guessing when evaluating generalization level. For example, the probability of simply guessing correctly on the receptive language probes is .25 (25%) because there are four choices presented. Although one could argue that any percentage greater than 25 reflects at least some generalization beyond chance levels, a more conservative approach is to calculate the probability of guessing correctly using the binomial statistic [31]. Using this method, we adopted a criterion of 67% correct responses (less than .05 probability that responses are attributable to chance guessing) for receptive targets, including expressively-taught, receptively-probed cross modal targets.

The probabilistic considerations for receptive to expressive probes are more complex, as the probability of guessing correctly is unknown and may vary from child to child, depending, at least in part, on the size of his or her individual lexicon. We adopted a 50% correct level to indicate that a word set taught receptively had generalized to the expressive modality. The rationale for this was that at least one of the words in the pair was learned fully, or that both words were learned partially in order to correctly name 50% of the expressive probes.

More broadly, success in generalization was defined as meeting criterion on the majority of legs of intervention (2 out of 3 legs). That is, the receptive to expressive generalization must have occurred in more than 50% of the probes in at least two out of three intervention legs to be counted as generalized from the receptive modality (treated) to the expressive modality (untreated). Similarly, at least two of the three legs must have shown at least 67% on the receptive generalization probes for an individual child to be credited with expressive (treated) to receptive (untreated) cross modal generalization.

Inter-observer agreement

IOA was determined for at least 33% of expressive and receptive probe conditions for each participant. IOA was conducted by an independent observer (another clinician or data analyst) by watching the session via videotape and recording the child’s responses as either correct, incorrect, or no response. Words produced with articulation errors were marked “correct” as long as the response was intelligible. The observer’s scoring was then compared to the treating clinician’s record of the child’s scores. The discrete nature of the probes made them readily identifiable to independent observation. IOA ranged from 95- 100% across sessions for all participants.

Procedural fidelity

Procedural fidelity was determined across 40-60% of intervention sessions and 40-60% of probe sessions for each participant. Procedural fidelity was measured by the independent observer either during the session or via videotape. The observer recorded the extent to which the intervention and probe condition trials were implemented as planned using a binary (implemented/not implemented) system. Implementation accuracy was calculated across each trial during intervention (20 trials) or across each discrete probe during probe sessions (48 trials).

The following items were recorded for each trial: clinician secured child’s attention, clinician delivered task directive correctly, response interval was the correct length, clinician praised testing behavior, appropriate corrective feedback was given and inter-trial interval was the correct length. Procedural fidelity ranged from 98-100% across sessions for all participants.

Plots of each child’s responses at baseline (cross modal and within-modality probes), during intervention (daily intervention probes) and follow-up probes (cross modal and within- modality probes) are displayed in Figures 1-5. Each participant met receptive and expressive criterion (80%, three consecutive sessions) for each word set, indicating that the target words were learned so that cross-modal generalization could be tested. That is, meeting criterion on these daily intervention probes was a prerequisite of the research questions. Participants required 7 to 35 sessions on an individual intervention tier to meet criterion.

Figure 1: Vocabulary learning and cross modal generalization: George.

Figure 2: Vocabulary learning and cross modal generalization: Nick.

Figure 3: Vocabulary learning and cross modal generalization: Sam.

Figure 4: Vocabulary learning and cross modal generalization: Joy.

Figure 5: Vocabulary learning and cross modal generalization: Tony.

Cross-modal generalization

Cross-modal generalization was higher for expressive-to-receptive than for receptive-to- expressive. For expressive-to-receptive cross modal probes, the criterion for success was set at 67%. All five children demonstrated cross modal generalization on the expressive-to-receptive probes for the majority of tiers of intervention (2 of 3; Table 2). For receptive-to-expressive cross modal probes, the criterion for success was set at 50%. In contrast to expressive-to-receptive probes, no children reached criterion for cross modal generalization on the receptive-to-expressive probes for the majority of tiers of intervention, with average receptive-to-expressive generalization at 17% and individual children’s averages ranging from 6% to 33%.

Table 2: Children’s success in probe condition immediately following intervention.

Secondary analysis: Maintenance

Maintenance probes were administered for the first and second legs of intervention. Proportions of correct responses on the first maintenance probe after intervention are shown in Table 3. All five children showed some maintenance within the trained modality for vocabulary targets, even several months later (within-modality probes after intervention). Children also showed high levels of maintenance in untaught modalities for which generalization occurred in the probe immediately following intervention (cross modal probes after intervention) using the same criteria as the cross modal probes. That is, when children generalized successfully immediately following intervention, they continued to be successful at subsequent time points. And when children were unsuccessful immediately following intervention (such as in the receptive-to-expressive probes), they were also unsuccessful at subsequent time points. Auxiliary Analysis: Control Target Sets.

Table 3: Children’s success in the probe condition during first maintenance probes.

As one would expect in real word studies, there were a few cases in which some untrained words were produced or comprehended. Each child was probed at each probe condition on a set of four control words. The criteria for success used during cross modal and general probes for the intervention sets were also used for the control set. One child, George, demonstrated modest evidence of learning for one control word in both the expressive and receptive modalities. No other children demonstrated evidence of incidental learning to criterion in either modality. Therefore, these results show intervention effects well above the observed levels of incidental learning.

Generalization is a ubiquitous problem both in clinical intervention research and in clinical practice, where learning in the clinic may not readily generalize to functional outcomes at home or in the classroom [32]. Many children with disabilities show poor performance on generalization tasks even when they perform exceptionally well on explicitly learned tasks [14,33]. The current study included an orthogonal test of expressive and receptive vocabulary learning wherein targets were taught independently with the express purpose of systematically testing cross modal generalization for vocabulary in children with DS. Broadly, cross modal generalization is often an assumed, but untested, outcome in clinical practice. Contrary to this assumption, the results of the current study indicate that cross modal generalization at the word level is not automatic in children with DS, particularly in the receptive to expressive direction. Although all children, including those with and without disabilities, show considerable variability in expressive and receptive vocabulary [34], it has long been established that children with typical language acquisition are capable of learning novel vocabulary in both modalities after only a few learning trials [19,20]. In contrast, the children in this study showed incomplete, partial cross modal generalization, particularly in the receptive-to-expressive generalization task, even after numerous training sessions and mastery in one modality across each replication.

Unlike previous studies of receptive and expressive vocabulary learning [11], the current study completely separated each language modality in order to examine the cross modal generalization. Vocabulary learning studies are often founded on a theoretical framework wherein expressive and receptive modalities are implicitly linked, or co-dependent, and that lexical storage is not complete unless the vocabulary item has been learned in both modalities [35]. Another common assumption is that words learned expressively will incidentally generalize to receptive knowledge as well, an assumption that is often valid in typically developing children [19]. In this study, children with DS displayed learning in both modalities (individually) and relatively high levels of expressive-to-receptive generalization, but they did not demonstrate the levels of expressive-to- receptive generalization that one would expect if the modalities were truly co-dependent.

The literature includes a number of reports that generalization across settings and stimuli is difficult for children with intellectual disabilities, particularly when explicit cues are not provided [36]. Although limited generalization occurs in most studies, generalization patterns are inconsistent and often quite low, which is consistent with cross modal generalization in the current report. Also, children with intellectual disabilities may have difficulty in the generalization of non-language tasks such as motor imitation in addition to language behaviors [36], suggesting that generalization is a more global problem extending beyond word learning. That is, difficulty in cross modal generalization in both expressive and receptive modalities is consistent an overall deficit in generalization abilities in children such as those with DS who display intellectual deficits.

Clinical implications

Clearly, the expressive-to-receptive and receptive-to-expressive generalization patterns have clinical implications. On the one hand, because expressive vocabulary successfully generalized to receptive vocabulary across all children in the study for the majority of legs of intervention, it may be useful to initially focus on expressive skills with the goal of also incidentally gaining receptive skills. But the results also indicate that clinicians should not assume that expressive learning will generalize for any particular vocabulary target or target set in children with DS, these assumptions must be tested systematically: The results of this study indicate that receptive vocabulary should be probed when words are initially taught expressively to determine whether cross modal generalization actually occurs in children with DS. In contrast, receptive vocabulary did not generalize to expressive modality in any of the children. This suggests that because receptive to expressive cross modal generalization is unlikely, clinicians should plan on augmenting receptive intervention with expressive teaching on vocabulary targets. Moreover, clinicians should assess cross-modal learning and provide additional instruction in the untrained modality as necessary to ensure full lexical acquisition. Finally, the current study also shows evidence of maintenance for learned vocabulary items over several months, which has not always been seen in the previous literature [11]. All five children showed relatively moderate levels of maintenance within the trained modality for vocabulary targets, even several months later. Children also demonstrated moderate levels of maintenance in untaught modalities for which generalization had occurred.

One limitation of the current study is the extensive probing required for comprehensive assessment of generalization in each probe session. During the intervention period, daily intervention probes were short, straightforward, and included only the vocabulary in which the child was being trained. During the cross modal and within-modality probe sessions, however, the inclusion of additional vocabulary (sixteen words as opposed to four words) and additional modalities (probed in both receptive and expressive modalities rather than one modality) may have resulted in some confusion and/or decreased motivation. Differences in the length and intensity of the probe conditions relative to the daily intervention probes may account for some of the variability in individual child responses.

Future research is needed on the relative independence of expressive and receptive modalities as these relate to vocabulary and language learning. The current study isolated receptive and expressive modalities. Although clinicians frequently target receptive language goals independently in activities such as following directions, there is an ongoing need for empirical data on receptive language intervention [9,10]. In addition, many clinicians may target expressive and receptive language concurrently, such as in storybook reading. Further research is needed to consider the relationship of language modalities when they are taught simultaneously and when they are taught in isolation. On the one hand, this may improve efficiency in terms of target delivery and cross modal generalization [35]; on the other hand, isolating modality may serve to highlight the target forms and result in at least some degree of incidental cross modal generalization. The current study suggests that an efficient approach of intervention may be to target expressive forms first, followed by additional receptive training for those words with little or no cross modal generalization. However, order of presentation should be studied systematically to determine whether there is indeed any advantage to simultaneous or sequential instruction of vocabulary across modality.

This research was completed with the support of award number R324A080143 Camarata (PI) from the Institute on Educational Sciences (IES) and by the Scottish Rite Foundation of Nashville. Gratitude is expressed to Joi Mitchell, who assisted with the vocabulary intervention sessions and to Will Camarata and Nick Bennet, who assisted with preparation of the training items.