Characterizing Caregiver–Child Interactions Through a Transactional Lens: A Baseline Analysis of a Caregiver-Implemented Intervention

Abstract

Purpose:

This study was motivated by the transactional model of development and examined the reciprocal influences that children and caregivers have on caregiver–child interactions (CCXs) prior to a caregiver-implemented intervention. We tested whether child communication characteristics were associated with caregiver strategy use and whether these strategies, in turn, influenced children's communication to understand how caregivers and children mutually shaped the language learning environment.

Method:

Caregiver–child dyads (N = 105) were participants in a randomized controlled trial. CCXs were collected when children were approximately 30 months of age, transcribed, and coded for four caregiver language facilitation strategies and child communication variables.

Results:

Least Absolute Shrinkage and Selection Operator regression and postselection inference indicated that child communication characteristics in CCXs were associated with both the frequency and type of strategies caregivers used. Children's overall communication acts were significantly associated with caregiver use of vocabulary strategies, whereas children's vocabulary diversity was significantly associated with caregiver use of sentence strategies. Mixed-effects logistic regression demonstrated that all four caregiver strategies significantly increased the likelihood of spontaneous lexical overlap in subsequent child turns.

Conclusions:

Prior to the intervention, caregivers and children reciprocally shaped the language environment. This supports a transactional perspective and warrants further consideration of reciprocal influences when assessing the impact of caregiver-implemented interventions.

Supplemental Material:

https://doi.org/10.23641/asha.32995796

Developmental language disorder (DLD) is a lifelong condition affecting children's language production and/or comprehension in the absence of other biomedical conditions (Bishop et al., 2017). By kindergarten, approximately 7%–11% of children will be diagnosed with DLD (Norbury et al., 2016). While language difficulties are a hallmark of DLD, this condition can also negatively impact children's educational outcomes (Duff et al., 2023), peer relationships (Toseeb et al., 2017), and mental health (Conti-Ramsden & Botting, 2008). Given its prevalence and broad developmental consequences, timely identification and targeted early intervention (EI) are essential.

Caregiver-implemented interventions are considered the gold standard for toddlers enrolled in EI, including those at risk for DLD (Finestack et al., 2022; Roberts et al., 2019; Roberts & Kaiser, 2011). In these interventions, EI providers coach caregivers to use language facilitation strategies during their daily routines. Numerous studies have shown that caregiver-implemented interventions lead to improvements in both caregiver use of intervention strategies and child communication outcomes. However, research on caregiver-implemented interventions has often treated children as passive recipients of these interventions. That is, most studies test a unidirectional path modeling the effects of the intervention (Buschmann et al., 2009; Girolametto et al., 1996; Roberts & Kaiser, 2012, 2015) and/or caregiver behavior (Crain-Thoreson & Dale, 1999; Girolametto et al., 1999; Hadley et al., 2017; Landry et al., 2012) on child outcomes. While this approach has yielded valuable information on the overall effects of caregiver-implemented interventions on children, the literature has largely overlooked the reciprocal influence children have on their caregivers.

The transactional model of development offers a framework better suited to capturing the complexity of caregiver–child interactions (CCXs) in caregiver-implemented interventions. This model asserts that children and caregivers exert reciprocal influences on one another, shaping their environment in real time (Sameroff & Chandler, 1975). These moment-to-moment exchanges, or transactions, not only shape immediate interaction but also influence future interactions and accumulate over time to impact children's developmental outcomes (Sameroff, 2009). Although the transactional model of development is frequently cited as a theoretical motivation for caregiver-implemented interventions (e.g., Adamson et al., 2020; Sone et al., 2021), the model's core tenets have not been operationalized and empirically tested in the context of intervention research. In general, a transactional approach prioritizes more proximal caregiver and child behaviors and considers the reciprocal influences they have on each other and the environment (Fogel, 2009; Sameroff, 2009). The current study takes a first step in filling these gaps, applying a transactional lens to baseline data from a randomized controlled trial involving 30-month-old children at risk for DLD and their caregivers.

The Relationship Between Child Communication and Caregiver Input

In caregiver-implemented interventions, the most immediate and intended change is caregiver use of language facilitation strategies (Brown & Woods, 2016). These strategies modify how caregivers socially engage with children and/or adjust the linguistic features of their input (Rowe & Snow, 2020). While EI coaching is a key support for caregivers to use language facilitation strategies (Brown & Woods, 2016), it is unlikely to be the only influence. From a transactional perspective, children's behaviors play a role in shaping how caregivers interact and talk to their children (Camarata & Yoder, 2002). It follows then that child communication characteristics would be linked with caregiver use of language facilitation strategies even before the intervention is delivered. Importantly, the responsive child-directed strategies examined in this study are not universal features of CCX. There is considerable cross-cultural variation in how caregivers engage with children and children develop the language(s) of their communities across a wide range of interactional contexts (Figueroa, 2024). Thus, the caregiver strategies examined in the current study should not be interpreted as necessary for language development. Rather, they represent a set of interactional and linguistic practices that align with specific sociocultural contexts, in this case, a sample of families in the United States who are primarily White, college educated, and middle class.

We explored baseline associations between child communication and four caregiver language facilitation strategies: imitation, expansion, responsive labeling, and responsive toy talk. All four are responsive strategies that modify different aspects of caregiver input. Imitations replicate some or all of the words children have just said (Levickis et al., 2014; Snow, 1981); expansions replicate all children's words and then add to it (MacWhinney, 2000; Snow, 1981); responsive labels name objects in the play environment (Girolametto et al., 1999; Harrington & Hadley, 2024; Mattie & Hadley, 2021); and responsive toy talk describes the state, qualities, or actions of objects in the environment using a noun subject (e.g., The box is open; The baby is sleeping; Hadley et al., 2017). Developmentally, imitations and responsive labels have been linked to children's early vocabulary growth (Girolametto et al., 2002; Levickis et al., 2014), whereas expansion and responsive toy talk support children's combinatorial speech, sentence development, and broader language outcomes (Girolametto et al., 2002; Hadley et al., 2017).

Prior work provides evidence that characteristics of children's communication are associated with features of caregiver input, including the specific strategies of interest. The rate at which children communicate is positively associated with the rate of caregiver speech (Hoff-Ginsberg, 1994), caregiver responsivity (Girolametto et al., 2002; Kuchirko et al., 2018; Vigil et al., 2005), and caregiver imitations (Flynn et al., 2004; Girolametto et al., 1999). The type of child communication is also important. Child gestures, especially pointing, have been linked to caregiver use of object labels (Olson & Masur, 2013). As children begin to talk more, their intelligibility (Frey et al., 2018; Su & Roberts, 2019) and increasing vocabulary (Girolametto et al., 2002) have been linked to specific features of caregiver input. For instance, in a sample of children with language delays, Girolametto et al. (2002) found that the child communication rate was positively associated with caregiver imitation but child vocabulary diversity was positively associated with caregiver expansions. These findings underscore that caregivers tend to adjust their input to align with their children's language level, with many studies demonstrating these patterns in the absence of intervention (Flynn et al., 2004; Girolametto et al., 2002; Olson & Masur, 2013). Although much of this work has focused on children with typical development, we anticipated that caregivers of children at risk for DLD would show similar patterns, aligning their baseline strategy use with the children's language abilities.

The Real-Time Impact of Caregiver Strategy Use on Child Communication

Caregiver language facilitation strategies are designed to positively impact child language. To date, this has been demonstrated by long-term outcomes. Only one caregiver-implemented intervention study has demonstrated the real-time effects caregiver strategies have on child communication. Brown and Woods (2016) found that toddlers were likely to communicate in response to language facilitation strategies, with the strongest effects observed for caregiver use of responsive strategies. These findings suggest that language facilitation strategies contribute to child communication behaviors in real time. The current study extends this work by focusing on a specific type of child communication: spontaneous lexical overlap.

Spontaneous lexical overlap is a relatively frequent, theoretically motivated, and clinically valued response, yet it is understudied in children at risk for DLD. Spontaneous lexical overlap is an umbrella term encompassing instances in which a child replicates at least part of their caregiver's utterance without being prompted to do so (Speidel & Nelson, 1989; Užgiris et al., 1989). This includes not only imitation (e.g., caregiver: “Big doggy”; child: “Big doggy”) but also broader instances, such as when children reuse only some words (e.g., caregiver: “The dog is eating”; child: “Dog food”) or reorder words (e.g., caregiver: “Do you think the doggy likes treats?”; child: “treat for doggy”). Lexical overlap is overtly transactional, as it reflects children's immediate responsiveness to their caregivers' language (Užgiris et al., 1989). It is widely accepted that children produce lexical overlap relatively frequently (Che & Brooks, 2021), and many accounts of child language development position lexical overlap as a useful communicative strategy (e.g., Snow, 1989). Clinically, verbal imitation is often considered an early developmental milestone (Kwok et al., 2018; Tamis-LeMonda et al., 2001), and infrequent imitation may be viewed as a risk factor (Kwok et al., 2018). Despite its theoretical and clinical relevance, only two studies have examined spontaneous lexical overlap or verbal imitation in children with language delays (Che et al., 2016; Kirchner & Prutting, 1987).

In this study, we examined children's spontaneous lexical overlap and characterized its association with caregiver language facilitation strategies. While no prior studies have investigated this in children at risk for DLD, research with typically developing children suggests that caregiver input shapes both the opportunity for and content of children's lexical overlap (Flynn et al., 2004; Užgiris et al., 1989). Most studies have focused on caregiver imitation and expansion, revealing that they are associated with higher rates of child lexical overlap (Casla et al., 2022; Masur & Rodemaker, 1999). Researchers have found that children are likely to respond immediately to caregiver imitations and expansions with lexical overlap (Clark & Bernicot, 2008; Farrar, 1987; Harrington et al., 2025; Masur & Olson, 2008; Scherer & Olswang, 1984). Like imitation and expansion, responsive labels and responsive toy talk present children with simple language input that is related to their focus of attention. Moreover, responsive labels place the noun in phrase- or sentence-final position, which is a highly salient position for subsequent child lexical overlap (Olson & Masur, 2012). To date, no studies have examined responsive labeling or responsive toy talk in relation to child lexical overlap.

Across the literature, we see evidence that specific types of caregiver input elicit contingent child responses. In addition to informing the types of caregiver input explored in the present study, these findings offer insight into how sequential caregiver–child turns, sometimes referred to as “serve and return,” may facilitate discourse (Houen et al., 2026). Through contingent responding, each caregiver and child turn shapes the next contribution to the interaction (Houen et al., 2026; Masek et al., 2021; Pickering & Garrod, 2004). These brief exchanges may serve as a foundation for longer multiturn conversations, which have been linked to later language development (Houen et al., 2026; Masek et al., 2021). Building on this perspective, the current study focused on a linguistically defined form of contingent responding—children's spontaneous lexical overlap. In line with existing literature, we predicted that caregiver language facilitation strategies would increase the likelihood of lexical overlap on children's next turn.

Purpose and Research Questions

The purpose of this study was to empirically test predictions from the transactional model of development by operationalizing and examining the reciprocal influences that children and caregivers have on interaction prior to a caregiver-implemented intervention. Specifically, we focused on key components of caregiver-implemented interventions, caregiver language facilitation strategies, and child communication behaviors and asked the following:

1. What child communication characteristics are associated with caregiver use of language facilitation strategies?

2. Do caregiver language facilitation strategies increase the likelihood of children's spontaneous lexical overlap on their subsequent turn?

Method

Data were obtained from the Maximizing Outcomes for Preschoolers with Developmental Language Disorder database (NCT03782493; Roberts et al., 2018). This randomized controlled trial was approved by the institutional review board at Northwestern University (STU00207758), with reliance agreements in place from Vanderbilt University and the University of Illinois Urbana-Champaign. Researchers from Northwestern University and Vanderbilt University recruited families by contacting EI providers, early education providers, and pediatrician offices and by posting local and online advertisements (Hadley et al., 2026). All families provided informed consent prior to participation. Data collection began in April 2019. Prior to the COVID-19 pandemic, 10 families completed the baseline visit in a research-based playroom at their corresponding site (Vanderbilt or Northwestern). All other families (90.7%) participated remotely in all aspects of the study. Detailed information about the clinical trial is available on the OSFHome website (Roberts et al., 2025).

Participants

Families were eligible for the clinical trial if children were between 30 and 31 months of age, one caregiver was willing to participate in caregiver training, English was the primary language spoken to the child at home, and children attempted to imitate at least 10 words via parental report (Roberts et al., 2025). Children also had to demonstrate (a) typical nonverbal cognitive abilities, defined as a T score of 35 or greater on the Visual Reception subscale of the Mullen Scales of Early Learning (Mullen VR; Mullen, 1995); (b) low to below-average receptive language abilities, defined as a standard score of 95 or less on the Auditory Comprehension subscale of the Preschool Language Scale–Fifth Edition (PLS-5; Zimmerman et al., 2011); and (c) an expressive language delay, defined as a standard score of 80 or less on the Expressive Communication subscale of the PLS-5 or scores below the 10th percentile on two of the three following sections of the MacArthur–Bates Communicative Development Inventories (Fenson et al., 2007): words produced (Part I, Section A), mean of the three longest sentences (Part II, Section D), and complexity (Part II, Section E).

Exclusion criteria included diagnosis of a neurodevelopmental disorder other than DLD; caregiver report of a major medical condition; hearing loss or hearing thresholds greater than 20 dB; or any problems chewing, sucking through a straw, or blowing bubbles. Children were screened for autism prior to baseline assessments using the TELE-ASD-PEDS (Corona et al., 2020) and excluded if their scores exceeded 10. Four investigator overrides were made. Four children who were unable to complete the administration of the Mullen VR (n = 2) or had scores just below the cutoff (n = 2) were included because assessors reported child noncompliance during test administration rather than inability to perform the items. While 108 families were enrolled, three participants were excluded from the analyses due to a later diagnosis of a neurodevelopmental disorder other than DLD. Therefore, the final sample was 105 caregiver–child dyads.

Tables 1 and 2 list the demographic characteristics of children and their caregivers, respectively. Caregivers were primarily biological mothers (92.4%) who identified as White (84.8%) and not Hispanic (87.6%). Over 75% of caregivers reported having a college degree. Children ranged in age from 29.63 to 32.79 months (M = 30.46 months, SD = 0.51); 59.3% were boys, and 40.7% were girls. The racial makeup of children was 82.9% White, 4.8% Black, 9.5% more than one race, and < 1% American Indian or Alaska Native. Children with Hispanic ethnicity made up 13.3% of the sample. Although about 65% of children were enrolled in community-based EI services, fewer than 20% of caregivers reported receiving coaching within their speech-language pathology or developmental therapy sessions (see Supplemental Material S1 for details on community-based services). For full details on the sample, refer to Hadley et al. (2026).

Table 1.

Caregiver demographics.

Variable	n (%)
Relationship to child
Adopted mother	4 (3.8)
Biological father	3 (2.9)
Biological mother	97 (92.4)
Other relation	1 (1.0)
Ethnicity
Hispanic	11 (10.5)
Not Hispanic	92 (87.6)
Not reported	2 (1.9)
Race
American Indian or Alaska Native	2 (1.9)
Asian	1 (1.0)
Black/African American	6 (5.7)
White/Caucasian	89 (84.8)
More than one race	4 (3.8)
Other	1 (1.0)
Not reported	2 (1.9)
Education
No college	8 (7.6)
Some college (no degree)	16 (15.2)
Specialized training/trade school	1 (1.0)
College graduate	42 (40.0)
Graduate/professional training	39 (37.1)
Annual family income
Greater than $249,999	11 (10.5)
$150,000–$249,999	12 (11.4)
$100,000–$149,999	19 (18.1)
$75,000–$99,999	21 (20.0)
$30,000–$74,999	29 (27.6)
Less than $30,000	3 (2.9)
Not reported	10 (9.5)

Table 2.

Child demographics.

Variable	n (%)
Sex
Female	44 (41.9)
Male	61 (58.1)
Ethnicity
Hispanic	14 (13.3)
Not Hispanic	90 (85.7)
Not reported	1 (1.0)
Race
American Indian or Alaska Native	1 (1.0)
Asian	0 (0.0)
Black/African American	5 (4.8)
White/Caucasian	87 (82.9)
More than one race	10 (9.5)
Other	1 (1.0)
Not reported	1 (1.0)
Enrolled in a community-based EI
Yes	68 (64.8)
No	37 (35.2)

Note. EI = early intervention.

Procedure

All families participated in multiple baseline assessment sessions, including collection of a CCX. For most families, the CCX was completed remotely, with families interacting in their homes and an assessor connected via a video communication platform. Caregivers were instructed to engage with and talk to their child as they would normally during a 5-min snack, 5 min of book reading, and 20 min of play. The research team provided all families with a standard set of books and toys to create similar opportunities for play and associated language use across the dyads. For the baseline CCX play section, toys included a car set, magnetic construction tiles, a birthday cake playset, stacking houses and animals, a pretend microwave and food set, safari truck set, and an animal puppet. Families were not required to play with all toys during the CCX. The research team sent families an iPad (cellular data enabled, if necessary), iPad case, tripod, tripod mount, and microphone. The CCX samples were audio- and video-recorded using an iPad and the embedded recording system of a video communication platform. A trained assessor was connected to facilitate the session and recording. Breaks and/or discontinuation were offered, if necessary. Discontinuation was rare but occurred during the snack portion for 13 dyads (12.4%) and in the book portion for one dyad (< 1%) at the baseline measurement point.

Transcription and Coding Procedure

A team of trained research assistants (RAs) at the University of Illinois transcribed and coded the CCX samples. All RAs at this site were naive to participant intervention status. RAs transcribed caregiver and child utterances using Systematic Analysis of Language Transcripts (SALT; Miller & Iglesias, 2017) conventions in a series of passes: adult transcription, child transcription, and consensus (Hadley et al., 2014). In the adult and child transcription pass, RAs transcribed all the utterances of the assigned speaker, including the child's nonverbal gestures (Mattie & Hadley, 2021). RAs transcribed conservatively, listening to an utterance a maximum of three times to avoid inflating estimates of speakers' intelligibility. During the consensus pass, an independent RA confirmed the original transcription and changed words or morphemes to unintelligible when they could not confirm them. Consensus transcribers also marked pauses of ≥ 3 s between caregiver utterances and between caregiver and child utterances (McDuffie & Yoder, 2010; Preza & Hadley, 2024).

RAs at the Illinois site were responsible for coding transcripts. Only fully intelligible and spontaneous utterances were coded. Child utterances were coded across the entire CCX. Caregiver utterances were coded in the middle 10 min of the play portion. In the first coding pass, coders slashed bound morphemes and contractions in line with SALT conventions and used a pipeout procedure to standardize lexical variations of the same word (e.g., ducky|duck; Hadley et al., 2014). In the second coding pass, RAs coded lexical verbs and sentence subjects following Hadley's (2019) coding scheme. A series of computerized searches were used to check codes for errors of commission and omission by both the original coder and an independent second coder.

In the third coding pass, RAs coded caregiver utterances for responsive, conceptual, and linguistic features using a multidimensional input coding scheme (Hadley & Preza, 2019). Of interest were the caregiver utterances coded as responsive labels and responsive toy talk. An utterance was considered responsive if it was semantically related to children's attention (Tamis-LeMonda et al., 2014) and well timed (i.e., in response to a child communicative turn or following a ≥ 3-s pause; McDuffie & Yoder, 2010; Preza & Hadley, 2024). An utterance was considered conceptually related to the environment if it referred to people, objects, and events that existed in the caregiver–child play environment. Two structurally defined language input variables, labels and simple active-declarative sentences, were coded when the responsive and conceptual criteria were also met. Responsive labels were operationally defined as single nouns, noun phrases, or sentences with the linguistic form pronoun + copula + noun phrase, and responsive toy talk was defined as simple active-declarative sentences with a noun subject. Exclusionary codes were used to identify active-declarative sentences that lacked any of these features. For examples and more details on exclusionary codes, see Hadley and Preza (2019). To establish coding reliability, 21 transcripts (20% of the data) were randomly selected and coded by an independent RA. Cohen's kappa (κ = .95) exceeded the a priori criterion of .80 (Sprent & Smeeton, 2001), indicating excellent levels of agreement.

Next, SALT transcripts were converted to Coding for Human Analysis Transcripts (CHAT) and coded for caregiver imitation, caregiver expansion, and child lexical overlap using the automatized CHIP command in Computerized Language Analysis (CLAN; MacWhinney, 2000; Sokolov & MacWhinney, 1990). Unique operators were added to the base CHIP command line (Harrington et al., 2025). A +q3 operator changed the default CHIP coding window from six utterances to two utterances. This ensured that instances of lexical overlap were temporally contingent on the communication partner's utterances and aligned with prior work demonstrating that most child overlap occurs within two caregiver utterances (Che et al., 2018; Scherer & Olswang, 1984). A −h operator excluded a predetermined list of conversational devices (e.g., yeah, okay; Sokolov & Moreton, 1994) from lexical overlap coding (e.g., CHIP ignored the overlap of yeah in Mother: Yeah; Child: Yeah). We also excluded a list of high-frequency function words (e.g., you, I, is) from child lexical overlap coding because they are less lexically meaningful than the overlap of content words (e.g., CHIP ignored the overlap of I in Mother: I feed the doggy; Child: I do it). However, high-frequency function words were not excluded from caregiver imitation and expansion coding because the addition of function words to make a child's previous utterance more grammatically complete is a recognized form of expansion (Child: dog eat; Mother: The dog is eating; Baker & Nelson, 1984). After CHIP coding, we manually inspected the CHAT transcripts for caregiver prompts. Using the FREQ command, we searched the caregiver tier of transcripts for utterances including “say” and “tell.” If caregivers explicitly prompted the child to use lexical overlap (e.g., Mother: say please; Mother: tell me dog) and the child responded using the prompted word (e.g., Child: please; Child: dog), we manually removed the CHIP codes from the child overlap tiers. This ensured that only spontaneous instances of child lexical overlap were coded. For further description of CHIP coding and a detailed appendix, see Harrington (2025).

Finally, we used the CHIPUTIL program in CLAN (MacWhinney, 2000) to link caregiver utterances and child responses based on the existing CHIP codes. If children responded to caregivers with lexical overlap (i.e., the child utterance was coded by CHIP), the CHIPTUIL program placed a $SOURCE code on the caregiver utterance. If children had not responded with lexical overlap or had not responded at all, the CHIPUTIL program placed a $NON code on the caregiver utterance. For more information on CHIPUTIL procedures, refer to Harrington et al. (2025) and Harrington (2025).

Measures

All measures were computed from the middle 10 min of play in the CCX because both caregiver and child utterances were coded in this portion of the transcripts.

Caregiver Variables

Caregiver strategies included imitation, expansion, responsive labels, and responsive toy talk (see Supplemental Material S2). Caregiver imitation was defined as a caregiver utterance that occurred within two utterances of the child and was an exact replication of all or some of the child's words (excluding conversational devices). Caregiver expansion was defined as a caregiver utterance that occurred within two utterances of the child and was an exact replication of all the child's words (excluding conversational devices) with new additions. Caregiver imitations were computed by totaling the number of $EXACT (i.e., only exact matches at the word level) and $REDUC (i.e., only exact matches and deletions at the word level) CHIP codes in the CHAT transcripts (MacWhinney, 2000). Caregiver expansions were computed by totaling the number of $EXPAN (i.e., only exact matches and additions at the word level) CHIP codes in the CHAT transcripts.

Responsive labels were defined as the responsive use of single nouns, noun phrases, or labeling sentences with the linguistic form pronoun + copula + noun phrase that referred to objects in the environment and was related to what the child was doing or saying (Hadley & Preza, 2019). Responsive toy talk sentences were operationally defined as the responsive use of an active-declarative sentence with a noun subject that referred to objects or events in the environment and was related to what the child was doing or saying (Hadley & Preza, 2019). Responsive labels and responsive toy talk were computed by exploring SALT transcripts for the [LAB] and [RD:TT] codes, respectively, and summing the total for each code.

Based on prior work linking certain caregiver strategies to specific child language abilities (Girolametto et al., 2002), we formed two caregiver strategy composite variables. The vocabulary strategy composite included the frequency of caregiver utterances containing a caregiver imitation or responsive label, separately or in combination, while the sentence strategy composite included the frequency of caregiver utterances containing caregiver expansion or responsive toy talk, separately or in combination. Because we suspected that the communication rate of caregivers would influence their strategy use, we computed the total number of caregiver utterances using SALT's automatic computation procedure.

Child Variables

We obtained child communication variables using SALT automatic computation procedures. Variables included the frequency of child communication acts, percentage of intelligible words, mean turn length in utterances, number of different words (NDW), and mean length of utterance in morphemes. We defined spontaneous child lexical overlap as the replication of at least one caregiver word, excluding conversational devices and high-frequency words from functional categories. It had to occur within two utterances of the caregiver utterance and could not be prompted. We allowed substitutions, deletions, and additions so long as the child replicated at least one nonexcluded word (e.g., Mother: that's where the little piece goes; Child: no, my piece). We computed spontaneous lexical overlap by totaling the number of $DIST codes on the %chi tier (MacWhinney, 2000) because CHIP includes this code for every instance of lexical overlap, regardless of type.

Sequential Caregiver–Child Variables

We computed sequential variables reflecting caregiver ➔ child turns by examining caregiver utterances for language facilitation strategy codes (i.e., $EXACT, $REDUC, $EXPAN, $LAB, and $RD:TT) and CHIPUTIL codes (i.e., $SOURCE and $NON). Because it was possible for utterances to contain two language facilitation strategy codes (e.g., Child: doggy; Caregiver: The doggy is running. $EXPAN $RD:TT), we needed to ensure mutually exclusive classification for our planned statistical analysis. Any utterance that met the criteria for caregiver imitation or caregiver expansion was counted in that respective category, regardless of whether it also met the operational definition for a responsive label or responsive toy talk. This resulted in the following sequential variables:

1. caregiver responsive label ➔ child spontaneous lexical overlap

2. caregiver responsive label ➔ no child spontaneous lexical overlap

3. caregiver responsive toy talk ➔ child spontaneous lexical overlap

4. caregiver responsive toy talk ➔ no child spontaneous lexical overlap

5. caregiver imitation ➔ child spontaneous lexical overlap

6. caregiver imitation ➔ no child spontaneous lexical overlap

7. caregiver expansion ➔ child spontaneous lexical overlap

8. caregiver expansion ➔ no child spontaneous lexical overlap

We repeated this process with nonstrategy caregiver utterances, or those that did not meet the operational definition for the four caregiver strategies, resulting in two additional variables:

1. nonstrategy caregiver utterance ➔ child spontaneous lexical overlap

2. nonstrategy caregiver utterance ➔ no child spontaneous lexical overlap

Together, these 10 variables characterized every partially and fully intelligible caregiver utterance by the type of caregiver utterance (i.e., responsive label, responsive toy talk, caregiver imitation, caregiver expansion, and nonstrategy) and subsequent child lexical overlap (i.e., $SOURCE or $NON).

Analysis Plan

Research Question 1 asked whether child communication characteristics were associated with caregiver language facilitation strategies. We entered candidate predictors into two separate least absolute shrinkage and selection operator (LASSO) regression models. The outcome for Model 1A was the caregiver vocabulary strategy composite variable (i.e., responsive labels and caregiver imitation), and the outcome for Model 1B was the caregiver sentence strategy composite variable (i.e., responsive toy talk and caregiver expansion). Visual inspection of the data revealed that both outcome variables exhibited a right skew that impacted the model fit. To address this, a square-root transformation was applied, yielding residuals that exhibited adequate normality and homoscedasticity.

In LASSO regression, the penalty term lambda (λ) shrinks predictor coefficients toward zero and eliminates less important predictors from the model (McNeish, 2015; Tibshirani, 1996). We used 10-fold cross-validation via the cv.glmnet function in R (Version 4.5.2; Friedman et al., 2010; R Core Team, 2021) to identify the value of λ that best fit the model, as measured by the mean square error. This function output the λ value that yielded the smallest mean square error (λ_min) and the largest value within 1 SE of λ_min (λ_1.se). We selected λ_1.se because it minimized mean square error while still applying a sufficiently large penalty term to yield a parsimonious set of predictors (Friedman et al., 2010; McNeish, 2015). LASSO models were refit with the entire data set, and coefficients were shrunk accordingly. Predictors that were not penalized to zero remained in the model and explained unique variance in the outcome variable. In a follow-up analysis, p values and 95% confidence intervals for each model's retained set of predictors were estimated using postselection inference via the fixedLassoInf function in R (Taylor & Tibshirani, 2016; Tibshirani et al., 2019). This method accounted for the selection bias introduced by the LASSO regression and adjusted the confidence intervals and p values accordingly.

Research Question 2 examined whether caregiver language facilitation strategies were associated with children's spontaneous lexical overlap. We conducted this analysis at the utterance level, where each utterance in the data set represented an observation. We restructured the data set from a participant-level format to a long-form utterance-level format using the dplyr (Wickham et al., 2023) and tidyr (Wickham et al., 2024) packages in R. Each of the 14,257 caregiver utterances served as their own row, with columns reflecting the participant ID, caregiver utterance type, and presence or absence of child spontaneous lexical overlap in response to that caregiver utterance. A mixed-effects logistic regression (MELR) model was fit using the lme4 package (Bates et al., 2015) in R via maximum likelihood estimation with Laplace approximation. The binary outcome variable was the presence or absence of spontaneous lexical overlap in each of the child's responses. Caregiver utterance type was modeled as a fixed effect with five levels: nonstrategy, caregiver imitation, caregiver expansion, responsive label, and responsive toy talk. The nonstrategy category was set as the reference level and modeled as the intercept. Participant ID was modeled as a random effect to account for nested data structure and subject-level variability. An analysis of variance test was used to compare the Akaike information criterion between the base model (with only the random effect) and the full model (with both random and fixed effects). Predicted probabilities and odds ratios were computed from the fixed-effects coefficient estimates.

Results

Research Question 1 asked whether child communication characteristics were associated with caregivers' use of language facilitation strategies. Table 3 summarizes the descriptive data for caregiver and child variables. Child predictor variables and caregiver communication acts (covariate) were entered into the two LASSO regression models (see Table 4). Model 1A retained the covariate and two predictor variables, yielding a mean square error (MSE) of 1.26. Postselection inference indicated that caregiver communication acts (covariate) and child communication acts were significantly associated with caregiver vocabulary strategies (all ps < .05), suggesting that children who communicated more often had caregivers who used more vocabulary strategies, when holding other variables in the model fixed. Model 1B retained the covariates and three predictor variables, yielding an MSE of 0.74. Postselection inference indicated that caregiver communication acts (covariate) and child NDW were significantly associated with the outcome (all ps < .05), suggesting that children with greater vocabulary diversity had caregivers who used more sentence strategies, when holding other variables in the model fixed. Together, these results support the hypothesis that child communication characteristics are associated with caregiver use of language facilitation strategies and that different child communication characteristics are associated with caregiver strategies geared toward different child language abilities.

Table 3.

Caregiver–child interaction descriptive data.

Variable	M	SD
Caregiver utterance types
Caregiver communication acts	146.54	46.39
Vocabulary strategy utterances	10.76	7.41
Caregiver imitation	4.19	4.34
Responsive label	8.28	6.60
Sentence strategy utterances	5.51	4.08
Caregiver expansion	4.66	3.75
Responsive toy talk	1.05	1.57
Child communication characteristics
Child communication acts	96.44	39.54
Child percent intelligible words (%)	49.25	21.31
Child mean length of utterance	1.43	0.40
Child number of different words	18.12	13.87
Child mean turn length	1.39	0.35
Child spontaneous lexical overlap	6.35	6.16

Table 4.

Least absolute shrinkage and selection operator regression and postselection inference output.

Variable	Model 1A vocabulary strategies				Model 1B sentence strategies
	Coefficient estimate	z score	p	95% CI	Coefficient estimate	z score	p	95% CI
(Intercept)	3.08				2.13
Child communication acts	0.03	2.16	.031	[0.03, 0.46]	0.05	1.51	.131	[−0.08, 0.33]
Child percent intelligible words	0				0.11	1.58	.115	[−0.08, 0.41]
Child mean length of utterance	0				0
Child number of different words	0				0.23	2.04	.041	[0.02, 0.58]
Child mean turn length	0				0
Child spontaneous lexical overlap	0.0001	1.44	.152	[−0.12, 0.37]	0
Caregiver communication actsa	0.03	3.24	.001	[0.16, 0.50]	0.04	2.60	.009	[0.07, 0.36]
Model parameters	Lambda (λ1.se) = 0.30 MSE = 1.26				Lambda (λ1.se) = 0.16 MSE = 0.74

Note. MSE = mean square error.

^aIncluded as a covariate in the model.

Research Question 2 asked whether caregiver language facilitation strategies increased the likelihood of spontaneous lexical overlap in children's subsequent turn. Overall, 92 children (88%) produced at least one instance of spontaneous lexical overlap in the CCX. Supplemental Material S3 displays the observed frequency of child responses containing spontaneous lexical overlap relative to each caregiver utterance type, collapsed across all participants. Two MELR models were fit to test the association between caregiver utterance type and child spontaneous lexical overlap. Relative to the base model that included only the random effect (i.e., participant ID), adding the fixed effect (i.e., caregiver utterance type) significantly improved model fit, χ²(4) = 153.05, p < .0001. Table 5 presents the final model estimates and fit statistics. The variance of the random effect was 1.014 (SD = 1.007), indicating that participants' unique random intercepts varied from the model intercept by approximately one unit. The fixed-effect coefficient estimates for all strategy utterances were positive and greater than 1, indicating that each caregiver strategy significantly increased the likelihood of subsequent child lexical overlap relative to nonstrategy utterances (all ps < .001). Predicted probabilities indicated that the likelihood of child lexical overlap was low following nonstrategy utterances (P = .025) and increased to approximately .08 following each caregiver strategy utterance. Consistent with this pattern, odds ratios indicated that child spontaneous lexical overlap responses were about three to four times more likely following caregiver strategy utterances than nonstrategy utterances. Overall, model certainty was the lowest for responsive toy talk estimates, as evidenced by large standard errors and wide confidence intervals. Nonetheless, caregiver language facilitation strategies appeared to positively and significantly influence children's lexical overlap responses on their next turn.

Table 5.

Fixed-effects output for mixed-effects logistic regression model.

Variable	Estimate	SE	z score	p value	P	P 95% CI	OR	OR 95% CI
(Intercept)	−3.67	0.118	−31.19	< .0001	.025	[.02, .03]
Caregiver imitation	1.34	0.150	8.93	< .0001	.088	[.07, .12]	3.81	[2.84, 5.11]
Caregiver expansion	1.24	0.150	8.29	< .0001	.081	[.06, .11]	3.45	[2.58, 4.65]
Responsive label	1.16	0.143	8.13	< .0001	.075	[.06, .10]	3.19	[2.41, 4.22]
Responsive toy talka	1.23	0.352	3.51	< .001	.080	[.04, .15]	3.44	[1.72, 6.85]
Model fit statistics: Akaike information criterion: 4,716.1; deviance: 4,704.1; residual degrees of freedom: 14,251

Note. P = predicted probability.

^aResponsive toy talk results should be interpreted with caution given the limited observations relative to other utterance types.

Discussion

The purpose of this study was to investigate CCXs prior to intervention using a transactional lens. We found that, during play interactions, language facilitation strategies were present in caregiver input at baseline and that different child communication characteristics were linked to different types of strategies. At the same time, all four caregiver strategies were more likely to elicit children's spontaneous lexical overlap responses than were nonstrategy utterances. Overall, these findings support the transactional model of development, suggesting that children are not just passive recipients of their caregivers' input but that caregivers and children mutually shape the language learning environment (Sameroff & Chandler, 1975).

Our first finding revealed that, prior to receiving a research-based intervention, caregiver strategies were already somewhat aligned with children's language level. Within this primarily White, college-educated sample, caregivers tended to use vocabulary strategies with children who communicated frequently and sentence strategies with children who produced a wider variety of words. These findings are consistent with existing observational work in similar sociocultural contexts, which has linked children's rate of communication with caregiver responsivity (Girolametto et al., 2002; Kuchirko et al., 2018; Vigil et al., 2005) and more precise child language measures with specific types of responsive strategies (Clark & Bernicot, 2008; Girolametto et al., 2002; Olson & Masur, 2012). In fact, our findings extend the work of Girolametto et al. (2002), who identified positive associations between the rate of child communication acts and caregiver imitations and between child vocabulary diversity and caregiver expansions in a sample of 20 caregiver–child dyads. Our substantially larger sample size allowed for more robust statistical models that controlled for related and potentially confounding variables, thus enhancing both the validity and interpretability of these earlier findings. Clinically, these findings underscore the value of gathering information from both caregivers and children at the onset of EI. By understanding caregivers' natural tendencies, EI providers can reinforce developmentally appropriate strategies that caregivers are already using and time the introduction of new strategies to children's developmental level.

Our second finding revealed that each of the four caregiver language facilitation strategies significantly increased the likelihood that children would respond with spontaneous lexical overlap in their next turn. Specifically, children were three to four times more likely to respond with lexical overlap following a caregiver strategy utterance than a nonstrategy utterance. We expected that caregiver strategies, particularly imitation and expansion, would be linked with children's lexical overlap based on prior work (e.g., Casla et al., 2022; Masur & Rodemaker, 1999). However, previous studies have largely focused on the overall rate of caregiver and child lexical overlap. Our sequential findings extend this by demonstrating that caregiver imitation and expansion promote children's immediate spontaneous lexical overlap. This aligns with the transactional model, providing evidence that caregiver behaviors have an immediate impact on the subsequent behavior of the child (Užgiris et al., 1989).

We also extended prior work by including two additional caregiver language facilitation strategies: responsive labeling and responsive toy talk. Because of the need for mutually exclusive caregiver strategies in this analysis, instances of responsive labeling and toy talk were nonimitative. That is, caregiver utterances that met the operational definition for a responsive label and met the operational definition for a caregiver imitation or expansion were coded as the imitative strategy (i.e., imitation or expansion) and excluded from the responsive label category. Similarly, responsive toy talk utterances that were also a caregiver expansion were coded as the latter. This coding decision was motivated by prior work demonstrating that caregiver imitation and expansion were associated with children's lexical overlap responses (Farrar, 1987; Harrington et al., 2025; Scherer & Olswang, 1984). Given this, we separated nonimitative strategies (responsive labeling, responsive toy talk) from imitative strategies (imitation, expansion) and found that responsive labels and responsive toy talk were associated with children's immediate lexical overlap responses. This suggested that features of caregiver input beyond imitation and expansion may also facilitate child lexical overlap. Responsive labels, in particular, were semantically aligned with children's interests; contained content words in salient, phrase-final positions (Echols & Newport, 1992; Olson & Masur, 2012); and, on average, were less than two words in length, making them simple input utterances for children to comprehend. Although a significant association was also found for responsive toy talk, the overall low frequency of responsive toy talk utterances limits our ability to interpret this finding with similar confidence. Nonetheless, this study adds to the small body of work examining spontaneous lexical overlap in children with language delays (Che et al., 2016; Kirchner & Prutting, 1987), demonstrating that responsive language facilitation strategies promote lexical overlap responses in children at risk for DLD. This is a notable finding for this population, as children with language delays may not have the expressive language abilities to contribute novel linguistic content to their interactions. Lexical overlap may serve as a communicative tool, allowing children to verbally participate in the interaction and sustain conversation in a natural and semantically related manner (Clark & Bernicot, 2008; Kirchner & Prutting, 1987; Pickering & Garrod, 2004; Snow, 1981; Užgiris et al., 1989).

Importantly, by examining precise sequential caregiver and child behaviors, we provide new insights into theoretically motivated and measurable transactions. An extensive body of research has demonstrated that caregiver input that is responsive, developmentally appropriate, and embedded in multiturn interactions is associated with children's later language development (Levickis et al., 2014; Masek et al., 2021; Tamis-LeMonda et al., 2001). However, it remains unclear how these types of transactions come to be. The present findings offer one potential mechanism in which specific types of caregiver input, in this case, developmentally appropriate, responsive vocabulary and sentence strategies, promote child lexical overlap during play. Although these transactions were relatively infrequent, they represent a linguistically precise instance of “serve and return,” where the linguistic content of each turn was semantically linked. By excluding conversational devices and high-frequency function words when coding for child lexical overlap, we could be more certain that the child repurposed meaningful content words from the caregiver's prior utterance in their response (Harrington, 2025). This is notable, as it provides caregivers with a clear semantic anchor for their next turn, making it easier to extend the interaction by building on the child's contribution (Houen et al., 2026). Although beyond the scope of the current article, it seems possible that the transactions observed during play may function as valuable in-the-moment language learning opportunities and hold promise as an entry point into cohesive multiturn discourse. The current methodology lays the foundation to investigate this transactionally motivated hypothesis in future studies.

Strengths, Limitations, and Future Directions

This study leveraged data from the baseline of a large clinical trial, offering several strengths. The large sample size supported sufficiently powered analyses, and the narrow age range of the children (30–31 months) eliminated the need to statistically control for age. At the same time, children showed considerable variability in expressive language. We viewed this as a strength of the study because heterogeneity is characteristic of children at risk for DLD; thus, our findings are more generalizable. Conversely, the sample was relatively homogeneous in race, ethnicity, and family income. We acknowledge that this is a notable limitation. Future research should explore whether the observed patterns hold in samples with greater cultural, linguistic, and socioeconomic diversity.

Several methodological decisions contributed to the reliability and ecological validity of the data. Most CCX sessions were video-recorded remotely, allowing children and caregivers to interact in familiar home environments. All dyads were provided with a standardized age-appropriate toy set, resulting in interactions that were comparable across dyads. However, the presence of a research team member may have affected how caregivers and children interacted. Additionally, the requirement for caregivers to set up devices to record the interaction may have introduced more disruptions and technical issues than in a controlled in-person environment. Nonetheless, the recordings yielded high-quality data from naturalistic interactions. Moreover, transcription and both manual and automated coding procedures were carried out with high levels of accuracy and excellent interrater reliability, further contributing to the quality of data.

Our study's cross-sectional design and transactional framework offer a novel perspective on CCX. By focusing on a single time point, we characterized precise behaviors of caregivers and children and examined both as outcomes. Although this was an important first step toward documenting reciprocal influences, each direction (i.e., child → caregiver; caregiver → child) was analyzed independently. Future work should use analytical approaches capable of addressing bidirectionality within a single model to truly address questions regarding reciprocal influences. Moreover, longitudinal work is needed to assess how child characteristics impact caregiver use of vocabulary and sentence strategies over time and whether developmentally aligning caregiver language facilitation strategies optimizes child language outcomes.

Finally, this study introduced a feasible theory-driven method to characterize linguistically defined caregiver–child transactions: a caregiver language facilitation strategy followed by a child spontaneous lexical overlap response. We hypothesize that these may serve as a meaningful entry point into multiturn interaction, thus holding potential to enhance children's language learning opportunities. A valuable next step would be to characterize these transactional sequences longitudinally and explore their association with extended multiturn interactions. From an intervention perspective, it will be important to explore whether these patterns occur more frequently in the intervention group than in the control group and test whether they contribute to both proximal and long-term child language outcomes. The clinical trial database, which includes seven measurement points across 18 months, is a valuable data set for addressing these future directions.

Conclusions

This study took a first step in empirically documenting key aspects of the transactional model of development within the context of a caregiver-implemented intervention. We provided preliminary evidence of reciprocal influences during play interactions, in which children's communication characteristics were associated with caregiver use of language facilitation strategies, and these strategies, in turn, elicited spontaneous lexical overlap in children. The methods and findings hold promise for understanding the specific ways in which caregivers and children mutually shape children's language learning environments. This can be leveraged to optimize the design and evaluation of future caregiver-implemented interventions.

Data Availability Statement

Caregiver–child interaction transcripts are available via CHILDES (https://talkbank.org/childes/access/Derived/EMT-SF.html). Data sets of caregiver- and child-level data generated during the clinical trial are available via Dryad (https://doi.org/10.5061/dryad.sj3tx96g9). Individuals using the clinical trial transcripts and/or data sets should cite the grant and key publications describing the database and measures in any resulting presentations and publications. Sequential utterance-level data from the current study are available from the first author upon reasonable request.

Funding Statement

This project was supported by Grant U01DC017135 (awarded to Ann Kaiser, Pamela A. Hadley, and Megan Y. Roberts) from the National Institute on Deafness and Other Communication Disorders and an Illinois Distinguished Fellowship (awarded to Emily K. Harrington) from the University of Illinois Graduate College. This study was conducted as part of the first author's doctoral dissertation. We thank Laura Mattie and Meaghan McKenna for their valuable feedback and Jeff Grauzer for providing statistical support.