Cross-Morpheme Generalization Using a Complexity Approach in School-Age Children

Abstract

Purpose

This single-case study examines a complexity approach to target selection in grammatical intervention in three children with varying levels of mastery of tense and agreement. Specifically, we examine whether targeting a complex tense and agreement grammatical structure (auxiliary BE in questions) leads to generalization to other less complex and related tense and agreement markers (auxiliary BE in declaratives, copula BE, third-person singular –s, and past tense –ed).

Method

Three children (all boys; aged 5;5–9;7 [years;months]) with deficits in morphosyntax were enrolled in a treatment program targeting a complex grammatical structure (auxiliary BE in questions) following collection of multiple baselines. Children's performance on the complex structure and related tense and agreement markers were tracked before, during, and after the intervention across three different tasks.

Results

Results show that, despite its grammatical complexity, the target was elicited in all three children with incomplete mastery of the tense and agreement system. Furthermore, all children demonstrated generalization to expressive language by increasing their mean length of utterance by approximately one morpheme during spontaneous language production following intervention. All children demonstrated individual patterns of generalization to other tense and agreement structures not targeted during intervention.

Conclusions

These functional changes observed following intervention set the stage for future controlled studies to establish a stronger cause–effect relation. Taken together, this study contributes to an emerging body of work showing that complex grammatical targets may be used in intervention earlier than previously thought.

Children with developmental language disorder (DLD; also known as specific language impairment; Bishop et al., 2017) have particular difficulty with verb morphology marking tense and agreement (Leonard, 1998). As a group, these children tend to omit tense and agreement morphemes at rates higher than other types of morphemes (Leonard et al., 1997; Rice & Wexler, 1996; Rice et al., 1995, 1998), and this difficulty persists into adolescence (Rice et al., 2009). Although there have been efforts to develop tense and agreement interventions (e.g., Leonard et al., 2004, 2006), there is currently no gold standard treatment for the tense and agreement deficits characteristic of this disorder. Therefore, innovation in treatment for these types of grammatical deficits is greatly needed.

An important consideration in intervention concerns target selection in planning. Much of the existing research in the area of grammatical intervention with children has manipulated the learning environment and the scaffolding provided during grammar interventions. This tells us how to provide therapy but does not inform what grammatical morpheme to target to maximize generalization and gains. There has been some limited work examining target selection in treatment of grammatical deficits in children. For example, Leonard et al. (2006) showed greater generalization to past tense –ed following treatment of third-person singular –s relative to treatment of BE (i.e., is, are, and was) in children. To be clear, the research literature has been careful to control for treatment targets in previous studies that examine intervention effects. These studies sometimes report how treatment of one structure leads to gains in another and how generalization takes place. However, theoretical claims that attempt to guide and describe “system-wide generalization” across the grammatical system using effective target selection are limited. The study presented here is the second phase of a project to develop a new intervention to support tense and agreement growth by maximizing generalization and, therefore, outcomes. This study adopts a complexity approach to treatment target selection (Complexity Approach to Treatment Efficacy [CATE]; Gierut, 2007; Thompson & Shapiro, 2007), which has been effectively employed in individuals with grammatical deficits (Levy & Friedmann, 2009; Owen Van Horne et al., 2018; Thompson & Shapiro, 2007; Thompson et al., 2003). The central aim of this study is to examine the effectiveness of such an approach in children with morphosyntax delays to examine cross-morpheme generalization.

Complexity in Grammatical Interventions: Target Selection

In a classic developmental approach to treatment of a language disorder, the intervention would target early developing structures and build up to targeting more complex, later-acquired structures. Complexity approaches differ from many prior extant interventions by first targeting more complex intervention targets and then looking for generalization “downward” to the less complex items. This approach originated in treatments for phonological disorders where complex targets were found to generalize to less complex targets and lead to a faster and more effective treatment (see Gierut, 2007). The CATE approach has been successfully used to support intervention for grammatical deficits in adults with aphasia (Thompson & Shapiro, 2007; Thompson et al., 2003). The rationale for this approach is grounded in learnability theory (for a thorough explanation, see Gierut, 2007). As Gierut (2007) argues, complexity supports language learning within a learnability framework. Learnability theory posits that a learner (in this case, a child) must be provided with the crucial input about the complex components of a language (i.e., positive evidence) so that they can learn the “full range” of allowable structures. As children are learning, they have only internalized a subset of this grammar. Children's emerging grammar can only be scaffolded if the child is provided with full positive evidence in the input so that the child can acquire and be exposed to all parts of the language. The idea is that, if we provide complex rather than simple data in the input to children, we allow young learners to infer and continue acquiring all aspects of the target grammar given the full breadth of evidence.

The few studies that have extended a grammatical CATE approach to children with DLD have been successful and prompt further inquiry. For example, recently, Owen Van Horne et al. (2018) used a CATE approach in the treatment of past tense –ed in children with DLD. In this study, complexity was defined based on the phonological complexity (e.g., verb stem ending) and semantic features (e.g., telicity) of the carrier main verbs used in the intervention. They found that training past tense –ed on verbs that were “harder” in terms of phonological and semantic complexity led to generalization to increased past tense –ed production on less complex verbs, whereas the reverse was not true. These findings provide emerging support for the CATE approach to treatment of tense and agreement markers (i.e., past tense –ed). Similarly, Aguilar et al. (2018) showed that grammatical treatment in children was more effective in highly variable verb contexts compared to contexts where fewer verbs were used. These studies together offer helpful information for the selection of targets based on phonological and semantic complexity (i.e., Van Horne et al., 2018) and input variability. The next step in this line of work is to directly examine the role of grammatical target selection in promoting cross-morpheme generalization.

DLD manifests as persistent difficulty in both language production and grammaticality judgments with all morphemes that mark tense and agreement. For this reason, in this study, we define complexity in the context of syntax in an attempt to directly target the underlying grammatical weakness in DLD across the tense and agreement system. Consequently, the goal of this intervention is to target a complex syntactic structure and examine generalization to other tense and agreement morphemes. We are interested in generalization to morphemes that share underlying syntactic features of tense and agreement and not generalization of a single morpheme to other word forms. Instead, the complex target is treated across a variety of word forms to support mastery and generalization to other related structures. Given that tense and agreement deficits are ubiquitous in English-speaking children with DLD, finding generalization to other grammatical forms regardless of the surface properties would be the hallmark of a truly effective intervention for these deficits.

In order to adopt a complexity approach to treatment, it is necessary that the less complex generalization structures are related to the more complex targeted structure in at least one dimension (see Gierut, 2007). As reviewed by Gierut (2007), complexity can be defined from epistemic, ontological, and functional perspectives. This means that a number of theories can be used to define complexity, and these definitions can be distinct and independent, although they sometimes overlap. In this study, we follow the theoretical approach of prior studies using the CATE approach for the treatment of grammatical deficits by defining complexity based on linguistic theories of thematic assignment (Levy & Friedmann, 2009; Thompson & Shapiro, 2007; Thompson et al., 2003). Thematic assignment helps the listener comprehend “who did what to whom” across sentence structures. As we describe below, the process of thematic assignment can vary in complexity based on syntactic argument structures. In what follows, we explain the rationale behind the hypothesis that treatment of auxiliary BE in yes/no questions (as in Sentence 1 below) will generalize to other tense and agreement structures with less complexity in thematic assignment (Sentences 2–6):

Complex target structure:

1. Is the girl reading the book? (auxiliary BE yes/no question)

Less complex generalization structures:

1. The girl is reading the book. (auxiliary BE declarative)

2. Is the boy tired? (copula BE yes/no questions)

3. The boy is tired. (copula BE declarative)

4. She reads a book. (third-person singular)

5. He picked strawberries this summer. (past tense –ed)

Thematic assignment is least complex in canonical sentences that present the subject, verb, and object in that order, such as in Sentences 2 and 4–6. In these sentences, the verb assigns the agent role to preceding noun phrase in the subject position and the theme role to noun phrase in the subsequent object position. However, in the auxiliary BE yes/no question in Sentence 1, thematic assignment requires additional processes that increase the complexity of the syntactic structure. In particular, unlike Sentences 2 and 4–6, the verb appears first in the sentence, whereas the subject and object positions appear later. One theory that has been proposed to describe movement of the verb from its original position to the beginning of the sentence suggests that this process is achieved through syntactic movement (see Figure 1). Under this account, this syntactic movement leaves behind a trace in the original position of the verb. The mechanism of thematic role assignment in a sentence like Sentence 1 requires several additional syntactic operations compared to canonical sentences such as Sentences 2 and 4–6 that make it much more complex. These additional operations include (a) movement and relocation of the verb and (b) assignment of a syntactic trace in the original position linked to the original verb.

Figure 1.

(A) Declarative auxiliary BE sentence. (B) Yes/no auxiliary BE question showing the additional level of syntactic movement required (in red) to create the question from the declarative form.

Unlike auxiliary BE, copula BE yes/no questions (as in Sentence 3) also require movement of the verb and a syntactic trace but do not require a second verb. Specifically, auxiliary BE requires coordination with another verbal morpheme (–ing). This means that the syntactic structure available for thematic assignment is more complex in yes/no questions with auxiliary BE than with copula BE given that the latter occurs with a single verb. That is, this structure requires additional coordination and movement. Indeed, Brown (1973) found that copula BE was acquired before auxiliary BE in typically developing children, further supporting the hypothesis that auxiliary BE is more complex than copula BE.

Although there remain open questions about generalization, the theory of syntactic movement and coordination indeed has psychological reality in both typical and atypical language learners (e.g., Love, 2007; Love & Swinney, 1996; Nicol et al., 1994; Thompson et al., 1998; Zurif et al., 1995) and has been used to successfully guide intervention in prior studies. Furthermore, the complexity of sentences with syntactic movement is a particularly difficult construction for children with DLD. A body of work has proposed that children with DLD and adults with aphasia have precise difficulty with this process of syntactic movement (Davies, 2002; Friedmann et al., 2006; Friedmann & Novogrodsky, 2004, 2007; Novogrodsky & Friedmann, 2006; van der Lely, 1994, 1998; van der Lely & Battell, 2003). These linguistic theories have also been extended to intervention showing that, in adults with aphasia, treatment on structures containing more instances of movement leads to generalization to structures with fewer instances of movement (Thompson & Shapiro, 2007; Thompson et al., 2003).

Research Questions

This study examines two research questions. The first asks whether production of tense and agreement structures changes from baseline levels with the introduction of an intervention targeting auxiliary BE in questions. Here, we expected to replicate the findings of a recent preliminary study that demonstrated gains in auxiliary BE in questions in a single case using the same intervention employed here (De Anda et al., 2019). An open question from the prior case study was whether the findings would replicate in children with differing levels of tense and agreement knowledge in particular, given that prior research indicates differential treatment effects based on the level of grammatical skill prior to intervention (Fey & Loeb, 2002). Thus, we sought to extend the prior findings from the case study to children with a wider range of tense and agreement knowledge.

The second question examines the pattern of change against predictions from CATE. Specifically, we examine cross-morpheme generalization across tense and agreement structures and whether it follows the expected outcomes. We expect that, by training a complex grammatical structure (i.e., auxiliary BE in questions), gains in less-complex related treatment structures would be observed (i.e., auxiliary BE in declaratives, copula BE, third-person singular –s, past tense –ed). That is, we expect that gains in the production of auxiliary BE in questions should generalize to tense and agreement morphology broadly, although there are several possibilities for the pattern of generalization as we outline below.

Predictions for Generalization

Although we can use theories of thematic assignment (e.g., syntactic movement and coordination) specifically to predict generalization following a CATE approach, it is not clear whether one structure or another will show greater generalization within and across children. That is, although we can predict generalization to copula BE, third-person singular –s, and past tense –ed, there are several possibilities as to which structure might show the greatest generalization. Here, we propose three possible hypotheses that describe the relative gains across generalization structures.

First, it is possible that generalization takes place as a function of shared surface forms and underlying function, such that copula BE will show greater generalization than third-person singular –s and past tense –ed because it shares the BE morpheme with the target intervention structure (auxiliary BE). Furthermore, copula BE questions share the syntactic movement operation and therefore may yield the greatest change over declarative forms, third-person singular –s, and past tense –ed.

Second, it is possible that children will take a developmental approach to filling in absent knowledge through generalization, such that they might generalize to other tense and agreement forms that share the tense and agreement function (such as third-person singular –s and past tense –ed) first before copula BE. Such an account would not be in opposition to a CATE account but would suggest that a complex grammatical target provides positive evidence to support continued developmental gains. For example, children may prefer to generalize to declarative forms before interrogative forms involving movement even if the target intervention structure uses an interrogative form, as in the current study. Indeed, prior studies have shown limited generalization from auxiliary BE to past tense –ed (Leonard et al., 2006), suggesting that generalization may be more likely to copula BE in this study than past tense –ed.

Third, it is possible that individual children will generate different patterns of generalization depending on the gains in the target intervention structure. Some children may prefer a developmental approach, whereas others may generalize based on shared function or form. Prior studies have shown differential responses to intervention in studies with children with DLD (i.e., Camarata et al., 1994). Consistent with learnability theory, children will likely show differential responses to the same complex input because each of them brings a different level of knowledge to the problem of language learning. That is, some children may already have some structures in place and therefore may prioritize learning of absent structures based on this initial state (Gierut, 2007). The single-case design of this study allows us to test these three possibilities by examining gains in grammatical knowledge at the individual level. Taken together, this study begins to examine the treatment outcomes of a novel intervention approach targeting the core grammatical deficits for English-speaking children with language delays and deficits.

Method

Participants

Three participants with a range of tense and agreement skills were recruited from a university speech-language pathology clinic for this study following the review of clinical records. Criteria for recruitment included (a) school-age children, (b) language impairment diagnosis, and (c) recent speech-language assessments indicating specific difficulties in tense and agreement marking. Following the careful review of records, three participants were identified who were interested in participating in this study. None of the three children were receiving any other language intervention during their participation in the study.

Participant DA (male, aged 9;7 [years;months]) had a diagnosis of moderate expressive and receptive language delay as well as mild articulation disorder at the start of the study. For approximately 6 months prior to the start of the current intervention, he had received speech-language intervention at the university clinic targeting speech production and past tense morphology in English, as well as pronouns and possessives in both Spanish and English, given that Spanish was the primary language spoken by his parents. In addition, he received services at school targeting story retell, speech production, written expression, and response to wh-questions (who, what, when, where, why, and how), which were all conducted in English. As a result, DA's dominant and functional language of expression was English, and he communicated with his parents and family in English even when spoken to in Spanish, which was confirmed by prior records, parent report, and direct observation. In addition, we administered the Test of Early Grammatical Impairment (TEGI; Rice & Wexler, 2001), a norm-referenced standardized assessment of grammar. The TEGI provides an Elicited Grammar Composite score, which averages production accuracy across four probes examining third-person singular –s, past tense, BE, and DO, respectively. TEGI scores are reported as percent correct of the items attempted within each category. These scores are then compared to criterion scores based on optimal sensitivity and specificity levels for each age to determine if the child is functioning like those children with or without a language disorder. For DA, the Elicited Grammar Composite score of 21.5 is very low given that he is the oldest of the three participants. Similarly, although DA is some months older than the age limit for the TEGI, he falls well below criterion across all probes and the composite score for children between the ages of 8;0 and 8;11 (93%), thereby further confirming delay in mastery of tense and agreement consistent with the review of records.

The second participant, PM (male, aged 8;11), was a monolingual English speaker who was homeschooled and received therapy at the university clinic since the age of 4;1. PM received a diagnosis of phonological disorder at the age of 4 years and DLD at the age of 8 years. Therapy goals included phonological processing, phonological awareness, and production of grammatical sentences. Clinic records indicated moderate language impairment as well as therapy-targeted speech production and pronoun use. This was further confirmed with the administration of the TEGI: PM did not reach age criterion scores across all four probes and received an Elicited Grammar Composite score of 59.3, which also fell below criterion (93%).

Lastly, participant JJ (male, aged 5;5) was also a monolingual English speaker who had attended the university clinic since for approximately 6 months prior to the beginning of this study. He did not receive additional services in the school setting. Participant JJ's records demonstrated specific deficits in tense and agreement. He had a history of age-appropriate articulation, receptive, and expressive skills but specific deficits in grammar, including delays in pronouns, past tense, prepositions, possessions, and omission of auxiliary and copula BE. The Elicited Grammar Composite score on the TEGI was 81.95 for JJ prior to the start of this study, which falls above criterion for his age (66%). However, JJ did not meet the age criterion for all probes (i.e., past tense –ed). Together, the three participants demonstrate a range of tense and agreement skills that affords an opportunity to examine how the present intervention will promote gains in children who vary in abilities prior to the start of intervention.

Study Design

Eliciting a complex target such as auxiliary BE in yes/no questions was successful in a recent case study employing CATE with a preschool child with tense and agreement delays (De Anda et al., 2019). The intervention was delivered in a play-based context through story vignettes to support the child's production of the complex target with cueing support. Although the case study established feasibility of the intervention approach, it remains unknown whether the treatment leads to effective generalization to other less-complex grammatical forms following the predictions from CATE, as described above. In addition, the previous study explored the intervention in the context of a child with relatively strong language skills and specific deficits in tense and agreement. As such, it remains unknown whether the intervention will support grammatical gains in children with a range of language deficits, including incomplete mastery of tense and agreement in English. This study is the next step in this line of inquiry.

In this study, we employ a single-case AB pre-experimental design in three different English-speaking children who all varied in their tense and agreement knowledge. As reviewed by Byiers et al. (2012), this type of design is best characterized by one that captures correlation between the independent and dependent variables but does not necessarily imply a causal relation. Given the relative novelty of the present intervention approach, this design therefore serves as an important building block for future and more rigorously controlled experimental studies that systematically vary grammatical target selection methods. That is, such a design provides an objective set of preliminary findings examining the effects of the intervention (Kazdin, 2011). Furthermore, there exists one published case study of the present intervention in a child with widespread deficits in tense and agreement structures, despite general language abilities within age expectations. As such, it is important to replicate and extend the feasibility of this approach to children with incomplete mastery of the tense and agreement system.

A central focus of this study is an examination of generalization from one structure to another. Therefore, the treatment design employed in the intervention featured several characteristics to support grammatical learning in young children with incomplete mastery of tense and agreement in English. The present intervention design builds on previous research by employing many of the same treatment delivery strategies to promote grammatical acquisition while using a less common complex intervention target. First, treatment was provided in relatively high dosage through play-based drills. This ensured that children had opportunities for practice with the target structure at a high rate. Prior studies using recasting as a method of intervention have shown that children with language disorders in particular do not seem to benefit from low densities unlike their typically developing peers (Proctor-Williams et al., 2001). Second, children's expressive language skills were specifically targeted through elicited production. Such an approach was favored over other techniques given that the intervention introduced a relatively novel and complex grammatical structure that involved syntactic movement likely absent in the repertoire of children with morphosyntactic deficits. Elicited production leads to more stable learning than modeling alone for auxiliary is and auxiliary inversion (Ellis Weismer & Murray-Branch, 1989). Indeed, more passive treatment approaches (e.g., recasting) have not been effective for gains in auxiliary structures (Fey & Loeb, 2002). These approaches may be less effective for encouraging production of a new form as speculated by Fey and Leob (2002). Third, given that the combination of evoked production and modeling produces more stable learning (Ellis Weismer & Murray-Branch, 1989), a cueing hierarchy was employed to provide feedback after each attempt at producing the target structure. Lastly, the treatment was delivered in a play context to encourage motivation and engagement by the children. Most importantly, the feasibility of this specific intervention was established in a prior case study (De Anda et al., 2019). Together, this body of work and direct evidence support the use of the intervention to examine generalization from the target structure across the expressive tense and agreement system.

As shown in Table 1, children participated in a total of 14 sessions made up of pre- and posttesting, baseline data collection, and intervention delivery (see the following sections for details on each session type). Sessions were conducted twice a week for a period of approximately 2 months. Following pretest data collection, three baselines were collected over the course of 2 weeks measuring the dependent variable of interest following recommended practices in single-case designs in communication sciences and disorders (e.g., Byiers et al., 2012). The purpose of collecting three baselines is to establish the children's performance on the outcome variables prior to the start of intervention. Importantly, stability in baselines demonstrates internal validity. A general benchmark of ±10% variation in baseline data establishes internal validity (Kratochwill, 2013; McReynolds & Kearns, 1983). Having established stability in baselines, any change found during the intervention indicates a relationship between outcomes and treatment.

Table 1.

Study design across sessions.

Session no.	Timeline
	1	2	3	4	5–13	14
Segment	Pretest	Baselines (A)			Intervention (B)	Posttest
Probe		✓	✓	✓	✓ a	✓

^a Probe collected on Session 9.

In the current study, the dependent variable of interest was production of tense and agreement markers across a variety of structures following intervention (i.e., the independent variable) that targets a more complex form (i.e., auxiliary BE in questions). In particular, we were interested in capturing gains in tense and agreement structures. As such, three probes were developed for this study to capture growth. Specifically, a spontaneous language sample, elicited production, and a sentence imitation task were collected in that order. Following baseline data collection, five intervention sessions delivered treatment on the target structure. The following session collected data on all three probes once more to capture gains midway through the intervention. Four additional sessions provided intervention for a total of nine intervention sessions. At the end of the intervention sessions, probe data were collected once more at posttest.

The sessions were conducted by a trained speech-language pathology clinical fellow. A second team of research assistants observed all sessions, following along with the same task protocols used by the clinician delivering the intervention to minimize deviations. This resulted in a high level of fidelity such that there were no deviations in materials and task order. Similarly, adherence to probes and treatment dosage was high (99.22% and 99.88%, respectively) such that the number of trials presented in probes and during the intervention was highly consistent across participants. Coders did not know the specific aims of the intervention. This also ensured that coding and intervention were completed independently.

Tense and Agreement Probes

A central question in this study concerns children's ability to generalize syntactic knowledge to other less complex but related tense and agreement structures. As such, and following best practice (Kratochwill et al., 2010), three probes were developed for this study to assess the dependent variable of interest (i.e., gains in tense and agreement structures). Specifically, in addition to gains in the target intervention structure (auxiliary BE in both singular and plural questions), we examined generalization to third-person singular –s, past tense –ed, copula BE (questions and declaratives in both singular and plural contexts), and auxiliary BE (declaratives in both singular and plural contexts). The probes assessed these grammatical production outcomes with varying levels of structured support and variability (e.g., the spontaneous production tasks allowed the child to rely on preferred grammar). Specifically, the grammatical structures were targeted in a sentence imitation probe (maximal structure support and minimal task demands), an elicited production probe (moderate structure support and task demands), and a spontaneous language sample (minimal structure support and maximal task demands). In this way, this study provides a nuanced and precise characterization of generalization gains in nontreated tense and agreement grammatical structures.

Sentence Imitation

Stimulus development. Stimuli for the sentence imitation task included 250 sentences, with 25 sentences written for each of the following target structures: auxiliary BE declarative singular, auxiliary BE declarative plural, auxiliary BE question singular, auxiliary BE question plural, copula BE declarative singular, copula BE declarative plural, copula BE question singular, copula BE question plural, third-person singular –s, and past tense –ed. Each sentence had the same general frame: (copula/ auxiliary) + determiner + singular/plural noun + (main verb) + (third-person singular –s/past tense –ed) + verb/prepositional phrase (e.g., The baby is crying; Is the teacher asking questions? The dad is pouring milk). Note that all sentences were short clauses, likely to be within the children's memory buffer capacity. Each sentence contained a unique main verb (except for the copula sentences, in which the copula BE served as the main verb), and all verbs were considered to be in the lexicon of 5-year-old children (Hall et al., 1984) to ensure familiarity by all children.

The 250 sentences were separated into five unique lists of 50 sentences, administered at each of the five probe data collection points: three at baseline as well as one at midintervention and one at posttest visit. Each grammatical structure appeared 5 times in each list, and structures were pseudorandomized.

Task administration. Participants were instructed to repeat each sentence exactly as they heard it immediately following presentation by the clinician. A training sentence using a nontargeted grammatical form was used to allow the participants to ensure they understood task demands and had a chance to practice before the test items were presented. No repetitions were allowed. Imitation attempts were transcribed online by a research assistant in the room and were recorded for later transcription and coding. Imitation attempts were transcribed off-line by a second research assistant. Online and off-line transcriptions were compared, and discrepancies were reconciled by consensus among the research assistants and the third author (A. A.).

Coding. The sentence imitation task-coding system was developed to examine the research questions of this study, namely, whether children made gains in the use of the target structure (auxiliary BE in questions) and whether there was generalization to other tense and agreement structures. To this end, each sentence imitation attempt was scored on two levels: (a) overall accuracy and tense and (b) agreement marker accuracy in obligatory contexts. Participants changed an item's structure during imitation (i.e., producing a declarative when the target was a question) an average of 4.9 items per 50-item list (SD = 4.7). To calculate percent correct imitation at each level, the denominator represented both correct and incorrect attempts for each structure, whereas the numerator represented the number of correct attempts. Although we presented five items for each structure, children sometimes did not attempt all five sentences, and as such, the denominator for percent correct varied. Coding was conducted twice by independent research assistants, and discrepancies were reconciled by consensus among the research assistants and the third author (A. A.). A subset (15%) was coded for agreement and indicated acceptable interrater reliability (87%).

Elicited Production

Stimulus development. The elicited production task paralleled the sentence imitation task in that there were five lists of 50 sentences, each containing five items for each of the 10 structures listed above. While the lists contained the same verbs as the sentence imitation task, unique sentences were written that followed the sentence frame provided above. Importantly, the lists were counterbalanced such that any participant never got the same verbs in the two tasks at one time point (i.e., on the same day).

This task was modeled after the TEGI Third-Person Singular and Past Tense subtests such that a visual cue was provided along with a specific elicitation from the clinician using a puppet (see Appendix A). For example, to elicit copula BE questions, the visual stimuli (i.e., a picture of two rabbits with a bandage wrapped on their arms) depicting the target sentence (i.e., “Are the rabbits hurt?”) was presented with the following elicitation cue: “The puppet wonders if the rabbits feel hurt. What could he ask?” Unlike the sentence imitation task, items for each structure were presented together given that prompts varied. For example, to elicit third-person singular –s, we provided an image depicting the target sentence (“The student cuts paper”) and provided a sentence frame such as “Here is a student. Tell me what the student does.” To elicit past tense, we presented a series of two pictures showing an action before and after completion (i.e., target sentence: “The man washed the car”) and then said, for example, “Here the boy is washing (point to first picture). Now he/she is done (point to second picture). Tell me what he did.” To elicit copula BE declaratives, we showed pictures of characters in different states to elicit a target (i.e., “The boys are hungry”) by prompting, for example, “Tell me about the boys.” To elicit auxiliary BE questions, we presented a set of images showing an action being completed (i.e., “The rabbit is walking outside”) and then said, for example, “Look at the rabbit here (point to first picture). Now he is done (point to second picture). The rabbit walked inside. Tell me what is happening with the rabbit here (point to the first picture again).” Finally, to elicit auxiliary BE in declaratives (i.e., “Are the girls hiding?”), a picture with an action in progress was presented with the following elicitation prompt: “The puppet wonders if the girls hide behind the desk. What could he ask?” For training, a practice trial was provided for each target structure to ensure that the child understood the task along with immediate feedback. Feedback was reserved for practice trials presented for each respective grammatical structure.

Task administration. After training, the clinician presented the participant with the visual and elicitation cues. No repetitions were allowed, and feedback was not provided. Elicitation attempts were transcribed online by a research assistant in the room and were recorded for later transcription and coding. Visual stimuli for each sentence structure were presented on a tablet as the clinician provided an elicitation cue. Elicitation attempts were transcribed off-line by a second research assistant.

Coding. Elicited production task scoring paralleled the sentence imitation task-coding procedures. Online and off-line transcriptions were compared, and discrepancies were reconciled by consensus among the research assistants and the third author (A. A.). A subset (15%) was coded for agreement and indicated acceptable interrater reliability (87%). The primary difference between the elicited production and the sentence imitation task was that, for the elicited production task, the participants had more flexibility with what structure they could produce, resulting in significantly more instances of elicitation attempts, including a different structure than the target compared to the sentence imitation task (M = 21.9, SD = 8.1), t(14) = 6.9, p < .001. For percent correct calculation and further analysis, any target structure with fewer than three attempts was excluded. This conservative measure ensures that we more reliably captured mastery of a grammatical structure.

Spontaneous Language Sample

Task administration. A play-based spontaneous language sample was elicited by the clinician using different age-appropriate toy sets (i.e., a car park with attendants, mechanics, and valet or a castle with knights and horses) for each session. The language samples were an average of 18.4 min in length (SD = 3.5) and had, on average, 168.4 (SD = 38.1) total utterances and 143.3 (SD = 31.9) complete and intelligible utterances. During the child-led free-play, the clinician provided minimal guidance following recommended practices for this elicitation context (Demuth, 1996; Eisenberg et al., 2001). That is, the child generally led the play sequence, and the clinician followed. The clinician provided a single prompt to elicit the child's attempt at each target structure (e.g., “Tell me about…,” such as “Tell me about what the knight did in the castle” to elicit past tense). However, all types of feedback were withheld (e.g., no recasting) to capture children's spontaneous production with minimal support.

Coding. All of the child's productions during the language samples were transcribed, coded, and analyzed using conventions from Systematic Analysis of Language Transcripts (Miller & Iglesias, 2012) by trained speech-language pathology master's students. Standard language sample measures, including number of utterances, mean length of utterance in words (MLU_w) and morphemes (MLU_m), and number of different words (NDW), were generated by Systematic Analysis of Language Transcripts. Accuracy of production of each target structure in an obligatory context was manually calculated. Similar to the elicited production task, any target structure with fewer than three attempts was excluded from further analysis. Two independent trained research assistants transcribed and coded each language sample. Discrepancies were reconciled by consensus among the research assistants, and all coding was checked for accuracy by the third author (A. A.). A subset of the language samples (15%) were coded to examine agreement for each participant. Interrater reliability was acceptable (M = 96.10%, range: 85%–100%).

Intervention

The intervention was designed to target a complex tense and agreement structure (auxiliary BE in questions) and was adopted from a previous study (De Anda et al., 2019). The treatment was embedded in a face-to-face play activity employing manipulatives across five story vignettes with each individual child in a one-on-one context. Each vignette presented 15 verbs, each used once in a singular and plural context for a total of 30 target questions in each treatment session. All but one of the five vignettes were presented twice during the nine intervention sessions, resulting in a high dosage of 270 opportunities for the production of target auxiliary BE questions in the intervention across all nine sessions. This high level of verb variability is consistent with prior research showing favorable grammatical outcomes compared to low-variability contexts (e.g., Aguilar et al., 2018).

To elicit each question, the child was instructed to ask a puppet about the scenes. See Appendix B for an example intervention script. An attempt at the target structure was scored as correct if auxiliary BE was accurately produced in an obligatory question context with the correct subject number (e.g., Is the girl crying?). Incorrect responses included (a) no attempt at the target structure, (b) use of the incorrect subject (e.g., Is the girl crying? for the target Are the girls crying?), (c) a statement with or without auxiliary BE (e.g., The girl (is) crying), or (d) a question with auxiliary BE omitted (e.g., The girl crying?). A cueing hierarchy was used following incorrect responses. Specifically, after an incorrect response, the request was repeated (i.e., Ask if the dogs/cat/monkey are/is ____ing). If a correct response was still not produced, a model was provided (i.e., I'll ask the puppet. Are/is the dogs/cat/monkey ____ing? Now you do it.). Finally, if a correct response was still not provided, the target structure was elicited via imitation (i.e., Say, are/is the dogs/cat/monkey _______ing?). Each intervention session lasted approximately 15–30 min depending on the level of cueing required to elicit the target structure.

Planned Analyses

The first research question asked whether participants made gains in producing the target structure, auxiliary BE in questions, during the intervention. Analyses examined changes in the amount of support needed for correct production of the target structure by summing across each intervention session the prompt level at which the target response was provided for each trial. For the second question, whether there was cross-morpheme generalization across tense and agreement structures, the first set of analyses examined broad changes in measures obtained from the spontaneous language sample pre- to postintervention. Next, we evaluated gains in the target and generalization structures. Following recommended practices with single-case designs, visual inspection of data was employed to assess whether there existed a functional relation between changes in children's production of tense and agreement markers as a function of the intervention. Additionally, we calculated effect sizes using standard mean difference with correction for continuity to corroborate visual inspection of the data. Standard mean difference is a routine calculation employed for single-case studies and is considered an analog to Cohen's d. It is preferred over many other measures of effect size as a nonregression technique because it conservatively measures learning effects, demonstrates relative ease in interpretation, and does not assume normality, equal variance, and independence of the data (Busk et al., 1992). Furthermore, this particular measure is widely used in assessing behavioral changes in clinical populations (e.g., Beeson & Robey, 2006; Gierut & Morrisette, 2011; Olive & Smith, 2005; Thompson et al., 2010). At present, there do not exist guidelines for interpreting the size of the effect into traditional “small,” “medium,” and “large” classifications for grammatical interventions in children that would allow us to compare to other studies. Even so, the effect size calculations offer a relative measure that affords comparison within the study. Therefore, the effect size calculations were used to compare children's performance with each other and across tasks.

Results

Results are organized to answer each of the two research questions in turn, presented for each participant separately. First, we examine changes in children's production of the target structure during intervention to replicate and extend the findings from a previous case study to children with a range of tense and agreement skills. Second, we present results on children's broad grammatical gains (measured by data from the spontaneous language sample) and generalization patterns to other tense and agreement structures (as measured by the probes). Appendix C provides a summary of scores on each probe before, during, and after intervention for each participant.

Child DA

Production of Target Structure During Intervention

Over the course of the intervention, DA showed a decrease in the number of prompts required to correctly produce the target response (see Figure 2). Compared to his peers, DA started the intervention requiring the most support to produce the target; he required all four prompts to correctly provide the target in 13 trials during the first intervention session. In addition, DA increased his rate of producing the target on the first attempt over the course of therapy. DA made the greatest gains relative to other participants given that he started at the lowest level of accuracy at baseline such that, by the final session, he provided the target on the first attempt in 28 of 30 trials.

Figure 2.

Level of prompting required to produce target structure (auxiliary BE in questions) during intervention.

Generalization

On the language sample (see Table 2), DA made gains across measures of utterance length (MLU_m = 3.14–4.02, MLU_w = 2.93–3.75) and vocabulary use (NDW = 125–171) over the course of intervention. This suggests broad gains in productive language.

Table 2.

Children's measures of productivity during the language sample.

Measure	DA		PM		JJ
	Pre	Post	Pre	Post	Pre	Post
Mean length of utterance in morphemes	3.14	4.02	4.38	5.19	3.28	4.52
Mean length of utterance in words	2.93	3.75	3.93	4.69	2.88	4.00
Number of different words produced	125	171	120	179	140	157

Visual inspection of DA's production of all tense and agreement structures across the three probes (sentence imitation, elicited production, and language sample) showed a relatively stable baseline period followed by an increase in accuracy postintervention, although baseline stability was less reliable during the language sample (see Figure 3 for specific results and a summary of findings in Table 3). For example, on both the sentence imitation and elicited production tasks, DA's accuracy was low before the start of intervention across baselines. Following the intervention, DA's accuracy crossed these pre-intervention thresholds demonstrating gains in production over time. Similarly, during the language sample, DA had an average tense marker accuracy of 45% prior to the start of therapy, which increased to 63% by the end of intervention despite the relatively unstable baseline. Taken together, this suggests a functional relation between the intervention and cross-morpheme gains in tense and agreement structures for the participant who started with the lowest accuracy. Furthermore, the possible floor effect on DA's performance on the sentence imitation and elicited production tasks does not change the interpretation of the findings, although it is possible that it attenuates the observed gains since the task's difficulty could potentially constrain DA's performance. That is, on these tasks, we can conclude with confidence a functional relation between the intervention and performance given the stable baseline period, followed by gains after treatment. However, we cannot be sure if the size of this effect is larger in reality as it may be attenuated by a possible floor effect.

Figure 3.

Change in overall accuracy on tense and agreement structures in each of the three probe tasks. Dotted line denotes start of intervention.

Table 3.

Summarized findings from Figure 4 showing gains and effect sizes per structure from pre- to postintervention across tasks for each participant.

	DA	PM	JJ	Effect size a
Auxiliary BE questions b	✓	✓	✓	0.88
Auxiliary BE declaratives c			✓	0.04
Copula BE questions c	✓		✓	0.66
Copula BE declaratives c	✓	✓		0.41
Third-person singular –s c		✓		0.12
Past tense –ed c		✓		0.51

Note. Check marks indicate that the grammatical structure exhibited gains in at least two of the three probe tasks (i.e., sentence imitation, elicited production, and spontaneous language sample) and suggest stronger evidence of change.

^a Standard mean difference.

^b Grammatical structure targeted during intervention.

^c Generalization structures not targeted in intervention.

DA demonstrated variable performance in each tense and agreement structure across all three tasks over time (i.e., Figure 4). Such variability is expected given the differences in task demands across tasks and the single-subject design of the study. However, the question remains whether there exists a signal (i.e., possible treatment effects) within this expected noise (i.e., natural variability). Indeed, DA's increases in production accuracy per structure were 3 times the degree of observed decreases (M _increase = 45.04%, M _decrease = 9.06%). Together, this suggests that gains in tense and agreement structure production accuracy following the start of intervention were greater than the degree of expected variability.

Figure 4.

Gains in each structure across each of the three probes for each participant. Data at pretest are averaged across all three baselines for ease of interpretation across the different sentence structures.

Lastly, we evaluated the specific pattern of gains in each of the generalization structures (see Table 3). First, DA showed gains across all tasks in the target structure (auxiliary BE in questions), thus corroborating the gains observed within the intervention. Second, DA also showed consistent gains (i.e., growth in more than one task) for copula BE in declaratives and questions. The pattern of generalization for DA suggests that gains were more likely to be observed for structures that shared both similar form and function with the target. Indeed, although gains were observed for forms sharing only function (e.g., third-person singular –s and past tense –ed), these were demonstrated only on one of three tasks and therefore showed less reliable gains. This is corroborated by an analysis of effect size (see Figure 5), which further examined the degree of observed change. Specifically, across tasks, DA demonstrated the largest effect sizes in grammatical structures overlapping in both form and function (auxiliary and copula BE in declaratives and questions). In summary, DA was able to produce the target structure, auxiliary BE in questions, during intervention with greater accuracy and independence. Following the intervention, DA showed the largest and most consistent generalization to tense and agreement structures that shared a surface form (i.e., is and are shared by auxiliary and copula BE), including processes of thematic assignment such as syntactic movement.

Figure 5.

Effect sizes (standard mean difference) across the target (auxiliary BE questions) and generalization structures for each participant.

Child PM

Production of Target Structure During Intervention

As shown in Figure 2, over the course of the intervention, PM generally showed a decrease in the number of prompts required to correctly produce the target structure indicating greater independence over time. PM required little support in the first session, providing the target on the first attempt in 20 of 30 trials. However, PM's performance was more variable across sessions compared to his peers. Nevertheless, by the final three intervention sessions, he provided the target structure on the first attempt in 28, 29, and 26 out of 30 trials, respectively, consistent with the prior published case study with a separate child and with DA's performance.

Generalization

First, we evaluated broad generalization in grammatical skills. PM showed gains similar to DA on the language sample (see Table 2). Specifically, he increased in measures of utterance length (MLU_m = 4.38–5.19, MLU_w = 3.96–4.69) and vocabulary use (NDW = 120–179).

Next, we evaluated change across the three probe tasks at baseline and following the start of intervention. Visual inspection of PM's performance on the probes showed a relatively stable baseline period, followed by an increase in accuracy postintervention for the elicited production task with less clear results evinced on the sentence imitation task and language sample (see Figure 3). Like DA, the greatest increase in performance was on the elicited production task, with accuracy improving by about 20%. On the language sample, PM showed highly variable baseline performance with accuracy fluctuating between 40% and 82% prior to the start of intervention. After the start of intervention, performance stabilized to a moderately high level of accuracy, around 70%. On the sentence imitation task, PM performed better and with more consistency, such that performance was relatively high and stable around 70%. Yet, there was no clear gain in tense and agreement accuracy in this task.

Despite the variability observed across tasks averaging across tense and agreement morphemes, PM showed gains in several specific structures across tasks (see Figure 4). Where decreases were observed, the slope of change was half of that observed in structures showing increases in production. That is, as a whole, PM's increases in production accuracy for a specific grammatical structure were twice the size of observed decreases (M _increase = 38.44%, M _decrease = 17.66%). In addition, the specific gains in tense and agreement were such that consistent growth (i.e., gains observed in more than one task) was obtained for the target structure (see Table 3). Consistent gains were also made on copula BE declaratives, third-person singular –s, and past tense –ed structures, with less reliable increases in accuracy for copula BE questions and no change for auxiliary BE in declaratives. Together, the results for PM suggest generalization from auxiliary BE in questions to structures that share function (i.e., third-person singular –s and past tense –ed) as well as to those that share both function and form and have less syntactic movement (auxiliary and copula BE). Indeed, effect sizes showed that the largest gains across tasks were observed for third-person singular –s and past tense –ed for PM (see Figure 5). These structures are indeed related to the target auxiliary BE in questions as they share a similar function (they describe tense and agreement) but do not require the same kind of complex syntactic movement for thematic assignment. Notably, the effect size for the target structure (auxiliary BE in questions) was much smaller compared to the change observed for the other two participants, which may have also led to the differences in the specific pattern of generalization observed across participants. We return to this comparison across participants in the discussion.

In summary, like DA, PM was able to produce the target structure with greater accuracy and independence over the course of the intervention. In addition, he showed generalization to other tense and agreement structures, although the exact pattern of generalization was notably different compared to that of DA, as we will discuss below.

Child JJ

Production of Target Structure During Intervention

As shown in Figure 2, JJ was relatively successful in producing the target structure after a single prompt throughout the intervention. Like DA and PM, over the course of intervention, JJ showed a decrease in the number of required prompts. At the first session, JJ provided the correct target form after the first prompt in 25 of 30 trials. The additional five trials required two or three prompts for correct production. By the final session, JJ was able to produce the target form (auxiliary BE in questions) after the first prompt in 29 of the 30 trials (97%). Together, this shows that the previous case study findings showing the feasibility of eliciting a complex grammatical structure extend to DA, PM, and JJ, all of whom began the study with varying levels of tense and agreement skills.

Generalization

We first examined JJ's pattern of broad generalization patterns following intervention. On the language sample, like the other two participants, JJ showed an increase from pretest to posttest in utterance length (MLU_m = 3.28–4.52, MLU_w = 2.88–4.00) and vocabulary use (NDW = 140–157), as shown in Table 2. These observations suggest changes in expressive language when measured broadly (i.e., by MLU).

Next, we examined generalization to the tense and agreement system across the three probe tasks. As shown in Figure 3, JJ demonstrated a stable baseline period followed by a sharp increase in tense and agreement production accuracy on the elicited production task. Specifically, JJ's accuracy increased from below 20% accuracy at baseline to 34% by the end of the intervention, suggesting a functional relation between the intervention and gains on the target and generalization tense and agreement structures of interest. On the sentence imitation task and language sample, performance was less clear due to instability at baseline and a possible ceiling effect precluding strong conclusions about spontaneous language and imitation gains on this task following intervention. Once again, the elicited production task seemed to provide the most reliable pattern of change in JJ, similar to DA and PM.

Following the examination of broad pattern of change across tense and agreement structures, we examined specific gains in each of the six structures of interest (i.e., auxiliary and copula BE in questions and declaratives, third-person singular –s, and past tense –ed). Like DA and PM, JJ showed increases in production accuracy for many grammatical structures across tasks (see Figure 4). Increases in production accuracy were twice the degree of observed decreases (M _increase = 18.39%, M _decrease = 8.01%), suggesting functional improvements following the start of intervention. JJ showed consistent gains (i.e., increases in two or more tasks) in the target structure, as well consistent generalization to auxiliary BE in declaratives and copula BE in questions (see Table 3). Gains in third-person singular –s were observed, but these were less reliable and observed only on a single task. Overall, the pattern of findings in JJ suggests that auxiliary BE generalized more robustly to structures that shared both form and function (auxiliary and copula BE) relative to those that shared only function (i.e., third-person singular –s), including those that involved syntactic movement with less coordination for thematic assignment (i.e., copula BE in questions) similar to DA. Examining effect size corroborates these findings (see Figure 5) in that less complex structures sharing form and function showed the greatest degree of change. Yet, the effect size for third-person singular –s was notable, although it did not surpass the gains made in auxiliary BE questions and declaratives and copula BE questions.

In summary, all three children replicated the findings of a previously published case study demonstrating the feasibility of eliciting a complex grammatical structure (auxiliary BE in questions). Here, we showed that the intervention was feasible in children with deficits in tense and agreement. Furthermore, we showed that all three children generalized their skill with syntactic movement and coordination to other less-complex tense and agreement forms, although the pattern of generalization differed for each child. We interpret findings across participants below.

Discussion

This study examined two main research questions. First, we asked whether production of tense and agreement structures changed from baseline levels following an intervention targeting auxiliary BE in questions to replicate and extend prior findings to a new group of children with varying levels of tense and agreement skill. The second question examined the pattern of change against predictions from CATE (Gierut, 2007; Thompson & Shapiro, 2007), specifically, cross-morpheme generalization to less-complex tense and agreement structures. In this study, we defined complexity in terms of thematic assignment processes of syntactic movement and coordination. We expected that treatment of auxiliary BE in questions (i.e., Is the dog running?) would generalize to other related tense and agreement structures with less-complex syntactic movement and coordination.

Changes in Production of Target Structure During Intervention

The first research question examined production of tense and agreement structures following the intervention. As expected, we replicated findings from a prior case study and showed that we could elicit auxiliary BE in questions with greater accuracy and independence (i.e., decreased prompting) over the course of the intervention. In this study, all participants showed a decrease in the number of prompts required to correctly produce the target response. Furthermore, all children increased in their rate of producing the target on the first attempt, when the least amount of support was available, in at least 80% of trials by the end of intervention. That all children made gains in production of the target structure with the intervention is consistent with the findings of our previous case study (De Anda et al., 2019), showing that this intervention successfully elicits a complex intervention target, in this case auxiliary BE in questions, earlier than previously thought. The present findings also extend those from the case study by showing that the intervention is feasible in children with a range of tense and agreement skills.

Overall Generalization Across Tense and Agreement Forms

A unique contribution of this study is our examination of gains within the tense and agreement system following a CATE approach. Recall that the second research question examined the pattern of generalization following the intervention. All three children showed improvements in productive language and vocabulary use in spontaneous language from pre- to postintervention. Most notably, children's productive language improved such that, across spontaneously produced utterances, utterance length increased by approximately one morpheme or word for all three children. While we cannot discount the possibility of changes due to maturation, our findings suggest that the complex grammatical intervention may have promoted gains in expressive spontaneous language given the gains we observed in tense and agreement morphosyntax, as we outline below.

Next, we examined the generalization patterns to the tense and agreement system. Indeed, generalization patterns demonstrated individual variability. Recall that, following predictions from CATE, we expected treatment of a more complex tense and agreement structure would generalize to other related and less-complex tense and agreement forms. Across all participants, gains in tense and agreement production accuracy were most clearly demonstrated on the elicited production task, where a stable baseline period (e.g., less than 10% variability) was followed by a clear increase in performance after the start of intervention. That is, the degree of change following the start of intervention was greater than the variability observed at baseline. Notably, although the task may have shown a possible floor effect, it does not change the conclusion that a functional relation between the intervention and children's grammatical gains was observed given the behavioral changes following a stable baseline period in all three participants. Instead, the possible floor effect may attenuate the size of the intervention effect given that it may limit the range of possible gains by reducing the lower bounds of children's performance. Nevertheless, the conclusion that a functional relation exists between the intervention and grammatical gains is supported and most clearly demonstrated on this task.

In addition, the multi-assessment method of this study capitalized on the differences in task demands, such that we replicated prior studies showing that elicited production tasks are more difficult than sentence imitation tasks, as expected (Räsänen et al., 2014). As others have concluded, sentence imitation tasks may yield lower error rates than elicited production given that the target structure is provided and can be kept in memory in imitation relative to elicitation. As Ambridge and Rowland (2013) suggest, sentence imitation tasks may be “less sensitive” than elicitation tasks in revealing production abilities. Importantly, we are not suggesting that sentence imitation tasks are not valid (a body of research demonstrates they are particularly difficult for children with DLD; Abel et al., 2015; Conti-Ramsden et al., 2001; Dalal & Loeb, 2005; Oetting et al., 2016; Smolík & Vávrů, 2014) but that elicitation tasks may be relatively more sensitive for examining expressive abilities as they more closely match the natural processes of language production. Visual inspection of child data showed gains following a stable baseline period in the elicited production, suggesting an intervention effect.

In addition, all children showed greater increases in production accuracy than decreases across tasks and tense and agreement forms, further supporting the conclusion that an intervention effect was observed. The majority of grammatical structures that participants produced across the three probe tasks (elicited production, sentence imitation, and language sample) also showed an increase in production accuracy. Together, these findings suggest that gains in production accuracy of tense and agreement forms following the start of intervention were more frequent and greater than the degree of expected variability at baseline. These are the gains that may be underlying the growth in MLU and productive language that we observed. Thus, although we cannot conclude a clear cause–effect relation in a single-subject study, we have enough evidence to suggest a functional relation between the intervention and subsequent increases in production accuracy across the tense and agreement system.

Relative Pattern of Generalization Across Tense and Agreement Forms

So far, we have described gains to the entire grammatical system through standard measures (i.e., MLU_m) and production accuracy across tense and agreement structures. Given that gains were observed in the tense and agreement system, we next examined the relative pattern of generalization from the target structure (auxiliary BE in questions) to other related forms. Recall that, although we anticipated that syntactic movement and coordination would predict generalization from auxiliary BE in questions to other tense and agreement forms, there remained several hypotheses about the pattern of relative change across generalization structures. Specifically, we posited three predictions. The first two predictions suggested that children as a group would follow the surface form or underlying function as a pattern of generalization, respectively. The third prediction posited that children would have individual patterns of generalization to least complex forms based on their proficiencies at baseline.

First, at the group level, we observed that the most consistent gain was in the target structure, auxiliary BE in questions, corroborating the gains on this same structure during intervention. In terms of generalization, all three children showed reliable gains (i.e., growth in at least two tasks from pre- to posttest) to other, less-complex tense and agreement structures (see Table 3). Notably, the pattern of generalization varied by individual, suggesting that there is no single generalization response to the intervention. Specifically, PM demonstrated reliable and consistent gains in all declarative forms, such that production accuracy increased for third-person singular –s, past tense –ed, and copula BE all in declarative forms from pre- to postintervention. This may more closely match a developmental approach to generalization in that PM generalized to tense and agreement structures that appear earlier rather than later in development. Conversely, DA showed reliable gains in copula BE across declarative and question forms, suggesting that DA's pattern of generalization may have been governed by overlap in both surface forms (i.e., BE in both auxiliary and copula) as well as syntactic movement (copula BE questions). Similarly, JJ's most reliable generalization was to forms that shared surface form and syntactic movement, auxiliary BE declaratives, and copula BE questions. These patterns suggest that, although all three children demonstrated generalization to less complex forms, the pattern of generalization across tense and agreement structures varied.

The individual pattern of generalization is further clarified by considering the relative accuracy gains with the target form as a function of effect sizes. As shown in Figure 3, both DA and JJ made the greatest gains in the production accuracy of the intervention target across probe tasks. For these children who made the largest improvements in the intervention structure specifically, gains in other structures sharing both form and function (i.e., auxiliary BE and copula BE) were generally more reliable than gains in structures sharing only function (i.e., third-person singular –s and past tense –ed). Specifically, it may be that the relatively large effect sizes in target gains helped DA and JJ with syntactic movement in particular, leading to reliable generalization to question forms (i.e., copula BE in questions; see Table 3). Conversely, for PM, gains in the intervention target were moderate compared to other tense and agreement morphemes. This may have therefore limited generalization to other question forms involving syntactic movement, such as copula BE questions. Indeed, gains in auxiliary BE in question form were significantly attenuated in PM compared to JJ and DA. As such, it follows that the pattern of generalization would differ for PM given less reliable mastery of the target structure outside the intervention context. Indeed, although notable gains were observed for copula BE, this was limited once more to declarative forms for PM. Furthermore, PM made relatively larger gains in third-person singular –s and past tense –ed compared to other structures and compared to other participants. Together, the pattern of findings suggests that generalization across the tense and agreement system may differ at the individual level depending on the amount of improvement with the complex intervention target specifically. Indeed, such individual patterns of change are often observed in treatment efficacy research examining intervention effects in children with DLD (i.e., Camarata et al., 1994). That is, children who demonstrate significant gains in the target presumably have internalized some level of syntactic movement and coordination processes to support thematic assignment. In this study, it was these children that ultimately generalized to other forms involving syntactic movement specifically (i.e., DA and JJ) in addition to other related tense and agreement declarative forms. Conversely, when gains in the target structure were less reliable, mastery of syntactic movement may be less reliable. Under such conditions, generalization can still occur but may be limited to the least complex forms (i.e., declaratives, as for PM).

That the pattern of generalization was contingent on children's gains in the target structure further provides support for the complexity account in this study. A frequency account alone is insufficient in explaining the observed findings. Specifically, although highly frequent exposure to BE in auxiliary contexts (as in the intervention target) may have supported the use of BE in copula contexts, it is not clear that a frequency account alone can explain gains in other tense and agreement structures with different surface forms, such as past tense and third-person singular –s. Recall that past tense –ed demonstrated greater gains than copula BE in declaratives across children. Notably, past tense –ed does not share surface forms with the intervention target, and as such, a frequency account of the input would not account for these findings. A frequency account alone might predict gains in BE across copula and auxiliary contexts given the significant exposure to BE forms in the intervention and absent or smaller gains to past tense –ed and third-person singular –s. Instead, it seems more likely that we observed gains across tense and agreement structures due to the underlying syntactic complexity consistent with prior research in adults with aphasia (Thompson & Shapiro, 2007; Thompson et al., 2003).

Clinical and Theoretical Implications

What does the current study add to our understanding of treatment in children with morphosyntactic challenges? First, the central question of this study concerned generalization from a complex grammatical target to other less complex forms in the tense and agreement system. The motivation was that this would inform our understanding of target selection given the large body of work examining treatment delivery. Here, we see that eliciting a relatively nonfrequent and complex grammatical target (auxiliary BE in questions) can indeed support growth in other tense and agreement morphosyntactic forms, although the pattern of generalization differed depending on the children's response to the intervention. This study therefore corroborates an emerging and growing area of research that suggests that targeting complex words and morphology may be an effective treatment strategy for children with language deficits (Owen Van Horne et al., 2018). It also corroborates findings showing that grammatical treatment in highly variable contexts is favorable given that children saw dozens of different verbs during the intervention (Aguilar et al., 2018). Recall that CATE-based grammatical intervention has been successful in adults with aphasia and with younger children as well (Levy & Friedmann, 2009; Thompson & Shapiro, 2007; Thompson et al., 2003). To our knowledge, this is one of the few studies to examine the utility of syntactic complexity in supporting language learning in children with tense and agreement deficits. These findings provide growing support for a dimension of complexity that can be employed in grammatical interventions. Future work must disentangle other sources of complexity in guiding generalization patterns. Recall that syntactic movement has been successful used in adults with aphasia (Thompson & Shapiro, 2007; Thompson et al., 2003). This study extends this to school-age children by showing that it is possible to elicit a grammatical structure with complex syntactic movement that promotes gains to less complex but related structures. This suggests that clinicians may break from developmental patterns and incorporate complex targets into treatment earlier than previously thought. However, although this study begins to inform the effectiveness of this approach, we are not yet sure of its efficiency. It is unknown how this treatment compares to other extant approaches, including those that take a more traditional developmental approach. Future studies must therefore compare treatment conditions to determine whether complex targets lead to a slower rate of change as compared to traditional stepwise approach that increases in complexity over acquisition. They must also consider varying the dimension of complexity itself. In the current study, complexity was defined in terms of thematic assignment, but this is just one of many possible definitions (Gierut, 2007). Importantly, however, change must be measured across the grammatical system. Complex targets are very likely harder to master than less complex ones. Even so, such an approach would be warranted if it maximized change to the entire grammatical system. The goal is not necessarily for a child to learn one grammatical target at a time but to select targets that promote the greatest change across the entire grammatical system and promote generalization. This distinction has been explored in the phonological domain, and indeed, although the complex phonological target is hardest to master, it confers the greatest amount of speech sound acquisition across the phonological system (Gierut, 2007). Future studies must therefore examine the degree to which different targets lead to system-wide change in the context of grammatical gains in children with morphosyntactic deficits.

Second, this study contributes to the principles of treatment delivery and measurement in tense and agreement in children. Recall that the treatment design provided varied and frequent opportunities to practice the target structure through elicited production in a play-based context with cued feedback. Thus, children were given a high dosage of opportunities to specifically practice the target structure each time with unique verbs during a relatively short period. In this context, we observed the greatest gains and largest effect size with the grammatical intervention target across children. At the individual level, all three children demonstrated accuracy scores greater than 70% following treatment. This is remarkable given that some studies employing similar implicit methods have shown less effectiveness. For example, in the intervention study by Leonard et al. (2004, 2006), only four of 25 children reached mastery after 96 sessions, and in Plante et al. (2014), three of nine children demonstrated mastery after 23 sessions. Thus, it is possible that implicit approaches may be helpful provided that they require language production in children, although gains may be maximized by combining explicit methods as well (i.e., Finestack, 2018). Additionally, it is possible that the high dosage and nature of the intervention supported language learning. In a more recent study using conversational recasting, Plante et al. (2019) showed that providing 24 doses of recasting with unique verbs yielded comparable gains whether they occurred over the course of 15 or 30 min. Here, we elicited 30 opportunities to directly produce the target structure across unique verbs in under 30 min across nine treatment sessions. As such, clinicians should carefully consider dosage schedules and variability in treatment stimuli, which also likely support effective outcomes. Taken together, the current findings imply that children can demonstrate gains even with a relatively complex intervention target (i.e., auxiliary BE in questions) under certain treatment conditions.

In terms of principles of measurement, this study demonstrates the importance of tracking system-wide change across a variety of tasks with varying demands. This dynamic assessment approach allows us to examine the child's grammatical skills in different contexts with varying levels of support and expectations. Here, we observed the largest effect sizes in the elicited production task relative to other tasks. Furthermore, in this study, we measured change across the tense and agreement system to examine generalization from the target across other related structures. Some prior studies have examined change in nontarget structures as mutually exclusive to changes in target structures. Instead, here, we suggest that generalization across related structures may indeed be expected and therefore maximized in the treatment of grammatical deficits. That is, treatment should consider the entire grammatical system and explicitly plan for a pattern of generalization to support long-term learning and gains. Although much has been written about this approach for treatment in the phonological domain, future work should directly compare different treatment approaches and examine their ability to maximize generalization and therefore change across grammar systems. This is of practical importance as maximizing treatment efficiency is critical in the face of increasing caseload sizes for speech-language pathologists and other service providers.

Limitations and Future Directions

This study employed a single-case design to explore whether the present intervention would be associated with functional changes in children's skill in tense and agreement structures. These findings are promising and support continued research. The conclusions of the present investigation are limited because they have not established a cause–effect relation between the observed changes following intervention. Therefore, the next step in this line of inquiry is to use a more rigorous experimental design to replicate and extend the present findings by including a comparison group for additional experimental control. This would be important to ensure a baseline comparison based on a different target selection approach. Future work should also control the individual variability of the participants given that this heterogeneity in this study precluded stronger conclusions.

This study also leaves open the possibility that the functional changes we observed may not generalize to children's contexts outside the clinic room, although gains in spontaneous language were demonstrated. Long-term maintenance in contexts outside therapy should be examined to provide recommendations for a dosage schedule that would promote grammatical gains that are sustained and meaningful. This future work would provide support for the efficacy and effectiveness of the intervention. As mentioned previously, equally important is an investigation of treatment efficiency in the context of CATE. A central tenet of the complexity approach is that treatment of a complex target promotes more system-wide change than a simpler structure. Therefore, future studies should compare change across the grammatical system by varying the levels of complexity in the treatment target.

Conclusions

Having established the feasibility of eliciting complex grammatical targets in a child with morphosyntactic delays in a previous study, the present investigation set out to extend this work by examining the treatment effects and outcomes across the grammatical system in a single-subject design. Three children with varying levels of tense and agreement skills were all able to produce the complex grammatical target with minimal cueing following the intervention. More importantly, all children demonstrated generalization to other tense and agreement structures, although generalization patterns differed across participants. Together, this study adds to a growing body of work examining the effects of a complexity approach to treatment efficacy in grammatical intervention for children with morphosyntactic delays.

Appendix A

Elicited Production Example Protocol

Structure	Target	Elicitation cue	Visual cue
Third person	The student cuts paper.	Here is a student. Tell me what the student does.
Past tense	The man washed the car.	Here the boy is washing (point to left picture). Now he/she is done (point to right picture). Tell me what he did.
Copula BE declaratives (singular and plural)	Tell me about the brothers.	The brothers are hungry.
Copula BE questions (singular and plural)	Are the rabbits hurt?	The puppet wonders if the rabbits feel hurt. What could he ask?
AuxiliaryBEdeclaratives (singular and plural)	The rabbit is walking inside.	Look at the rabbit here (point to first picture). Now he is done (point to second picture). The rabbit walked inside. Tell me what is happening with the rabbit here.
Auxiliary BE questions (singular and plural)	Are the girls hiding?	The puppet wonders if the girls hide behind the desk. What could he ask?

Appendix B

Example Intervention Script

INTRO:

We're going to play a game. In this game, we have two dogs, a cat, a monkey, and a puppet. We can't talk to the animals because they speak a silly language that we don't know. They can't talk to us or understand us either because they don't speak English, like us. But the puppet speaks English and speaks the same silly language as the animals so we can talk to the puppet and the puppet can talk to the animals. So, if we want to know things about the animals, we have to ask the puppet.

Let me show you. I'll ask the puppet a question:

Examiner (E): “Can you jump up and down?”

Puppet (P): “Yes, see?” (puppet jumps up and down)

TRAINING: Now you ask the puppet a question. Ask the puppet if he can talk:

Target (T): can you talk?

See? We can talk to the puppet and he can understand us and answer us. So, if we want to know things about the animals, we can ask the puppet and he'll ask them because he can speak the same silly language as them and we can't. Do you understand?

PROMPTING:

Prompt on

Omission of AUX verb

Wrong form of AUX verb

No attempt at AUX verb (i.e., changes sentence structure)

No response

DO NOT prompt on

Incorrect subject number*

Changed main verb

Prompt levels

1. Repetition of request: “Ask if the dogs/cat/monkey are/is ____ing.”

2. Provide model: “I'll ask the puppet. Are/is the dogs/cat/monkey ____ing? Now you do it.”

3. Imitation: “Say, are/is the dogs/cat/monkey _______ing?”

SCRIPT:

1. (dogs stand in front of some form of food)

E: I wonder if the dogs are cooking the food. Ask if the dogs are cooking.

T: Are the dogs cooking?

P: Yes.

2. (cat moves to the food)

E: And I wonder if the cat is cooking the food. Ask if the cat is cooking.

T: Is the cat cooking?

P: Yes.

3. E: All done cooking! Now time to get the food ready to eat.

(dogs make cutting movements)

E: I wonder if the dogs are cutting the food. Ask if the dogs are cutting the food.

T: Are the dogs cutting the food?

P: Yes. They have to cut it first.

4. E: The cat wants to eat, too.

(cat makes cutting movement)

E: I wonder if the cat is cutting the food. Ask if the cat is cutting the food.

T: Is the cat cutting the food?

P: Yes. He has to cut it first, too.

5. E: Now I think they can eat!

(eating sounds)

E: I wonder if the dogs are eating the food. Ask if the dogs are eating.

T: Are the dogs eating?

P: Yes.

6. E: And I wonder if the cat is eating the food. Ask if the cat is eating.

T: Is the cat eating?

P: Yes.

7. E: How do they eat? Do they eat like people? I don't think so. I think they lick their food first.

(make licking sounds)

E: I wonder if the dogs are licking the food. Ask if the dogs are licking the food.

T: Are the dogs licking the food?

P: Yes.

8. E: Maybe the cat does the same thing. I wonder if the cat is licking the food. Ask if the cat is licking the food.

T: Is the cat licking the food?

P: Yes.

9. E: What do they do after they lick their food? I think they bite it.

(biting sounds)

E: I wonder if the dogs are biting their food. Ask if the dogs are biting their food.

T: Are the dogs biting their food?

P: Yes.

10. E: Maybe the cat does the same thing. I wonder if the cat is biting his food. Ask if the cat is biting his food.

T: Is the cat biting his food?

P: Yes.

11. E: Okay! They all bit their food. What happens next?

(chewing sounds)

E: I wonder if the dogs are chewing the food. Ask if the dogs are chewing.

T: Are the dogs chewing?

P: Yes.

12. E: Maybe the cat does the same thing. I wonder if the cat is chewing the food. Ask if the cat is chewing.

T: Is the cat chewing?

P: Yes.

13. E: Finally! All the animals finish eating. Wait! What about that monkey? He hasn't eaten. He doesn't know if he likes it.

(food moves to the monkey)

E: I wonder if he's tasting his food. Ask if the monkey is tasting his food

T: Is the monkey tasting his food?

P: No.

14. E: Oh no! Maybe the dogs can show him that it tastes good. I wonder if the dogs are tasting his food. Ask if the dogs are tasting his food.

T: Are the dogs tasting his food?

P: Yes.

15. E: Maybe the monkey can't eat on his own.

(dogs feed the monkey)

E: I wonder if the dogs are feeding the monkey. Ask if the dogs are feeding the monkey.

T: Are the dogs feeding the monkey?

P: Yes.

16. E: The cat probably has to help out too. I wonder if the cat is feeding the monkey. Ask if the cat is feeding the monkey.

T: Is the cat feeding the monkey?

P: Yes.

17. E: Good! Everybody has been fed. They probably feel thirsty now.

(animals move toward the drink. make drinking sounds)

E: I wonder if the dogs are drinking the water. Ask if the dogs are drinking.

T: Are the dogs drinking?

P: Yes.

18. E: The cat probably feels thirsty too. I wonder if the cat is drinking the water. Ask if the cat is drinking.

T: Is the cat drinking?

P: Yes.

19. E: Good! Now everybody can help clean up. Time to dump the leftover food in the trash. I wonder if the dogs are dumping their food. Ask if the dogs are dumping their food.

T: Are the dogs dumping their food?

P: Yes.

20. E: I wonder if the cat is dumping his food too. Ask if the cat is dumping his food.

T: Is the cat dumping his food?

P: Yes.

21. E: Whew! What a mess. Everyone needs to get to work.

(animals stand up and move around)

E: I wonder if the dogs are working on the mess. Ask if the dogs are working.

T: Are the dogs working?

P: No.

22. E: And I wonder if the cat is working on the mess. Ask if the cat is working.

T: Is the cat working?

P: No.

23. E: Oh no! Everyone needs to help clean. I wonder if the dogs are cleaning the mess now. Ask if the dogs are cleaning.

T: Are the dogs cleaning?

P: Yes.

24. E: And I wonder if the cat is cleaning the mess now. Ask if the cat is cleaning.

T: Is the cat cleaning?

P: Yes.

25. E: What kind of cleaning do they do?

(animals act like they're mopping)

E: I wonder if the dogs are mopping the floor. Ask if the dogs are mopping.

T: Are the dogs mopping?

P: Yes.

26. E: I think the cat does the same thing. I wonder if the cat is mopping the floor. Ask if the cat is mopping.

T: Is the cat mopping?

P: Yes.

27. E: What a lot of work for that cat. I'm not sure he can do it all by himself.

(dogs move over to where the cat is)

E: I wonder if the dogs are helping the cat. Ask if the dogs are helping.

T: Are the dogs helping?

P: Yes.

28. E: That cat still has a hard time.

(monkey moves over to where the cat is)

E: I wonder if the monkey is helping the cat. Ask if the monkey is helping.

T: Is the monkey helping?

P: Yes.

29. (animals stop moving)

E: Almost done cleaning! I wonder if the dogs are finishing the job. Ask if the dogs are finishing.

T: Are the dogs finishing?

P: Yes.

30. E: And I wonder if the cat is finishing the job. Ask if the cat is finishing.

T: Is the cat finishing?

P: Yes.

E: Everybody finishes! The End.

Appendix C

Performance on Each Probe Task Before, During, and After Intervention for Each Participant

Elicited production
Participant	DA			PM			JJ
Time point	Pre	Mid	Post	Pre	Mid	Post	Pre	Mid	Post
Auxiliary declaratives	0.75	0.77	0.73	0.45	0.13	0.27	0.91	0.94	1.00
Auxiliary questions	0.00	0.70	0.70	0.00	0.86	0.91	0.67	1.00	1.00
3s	0.31	0.00	0.25	0.69	0.86	1.00	1.00		1.00
Past –ed	0.30		0.20	0.33	0.60	1.00	0.00		0.00
Copula declaratives	0.87	0.71	0.67	0.92	1.00	1.00	1.00	1.00	1.00
Copula questions	0.00	1.00	0.67	0.00		0.57	0.00		1.00
Language sample
Participant	DA			PM			JJ
Time point	Pre	Mid	Post	Pre	Mid	Post	Pre	Mid	Post
Auxiliary declaratives	0.69	0.75	0.80	0.71	0.63	0.48	0.53	0.75	0.80
Auxiliary questions	0.50	0.80	0.91	1.00	0.50	0.57	0.78	1.00	1.00
3s	0.02	0.00	0.07	0.45	0.60	0.29	0.96	0.63	0.91
Past –ed	0.00	0.00	0.20	0.11	0.50	0.86	0.53	0.75	0.40
Copula declaratives	0.44	0.67	1.00	0.83	0.94	1.00	0.86	0.79	0.89
Copula questions	0.81	0.88	0.80	1.00	0.86	1.00	0.79	0.94	0.94
Sentence imitation
Participant	DA			PM			JJ
Time point	Pre	Mid	Post	Pre	Mid	Post	Pre	Mid	Post
Auxiliary declaratives	0.60	0.60	0.60	0.83	0.50	0.70	0.87	1.00	1.00
Auxiliary questions	0.07	0.30	0.70	0.67	0.40	0.90	0.90	0.90	1.00
3s	0.33	0.20	0.20	0.80	1.00	1.00	0.80	1.00	1.00
Past –ed	0.40	0.20	0.20	0.67	0.20	0.80	0.80	1.00	0.80
Copula declaratives	0.50	0.70	0.60	0.83	1.00	0.80	0.90	0.80	0.90
Copula questions	0.13	0.30	0.60	0.87	0.70	0.80	0.97	1.00	0.90

Note. Pre time point includes average of all baseline measures. Blanks in the table represent less than three attempts at production of the grammatical structure.

Funding Statement

This research was supported in part by the National Institutes of Health (Grant F31HD081933) to the first author and does not necessarily represent the views of the National Institutes of Health.